1 /* 2 * net/sched/sch_prio.c Simple 3-band priority "scheduler". 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 * Fixes: 19990609: J Hadi Salim <hadi@nortelnetworks.com>: 11 * Init -- EINVAL when opt undefined 12 */ 13 14 #include <linux/module.h> 15 #include <linux/slab.h> 16 #include <linux/types.h> 17 #include <linux/kernel.h> 18 #include <linux/string.h> 19 #include <linux/errno.h> 20 #include <linux/skbuff.h> 21 #include <net/netlink.h> 22 #include <net/pkt_sched.h> 23 #include <net/pkt_cls.h> 24 25 struct prio_sched_data { 26 int bands; 27 struct tcf_proto __rcu *filter_list; 28 struct tcf_block *block; 29 u8 prio2band[TC_PRIO_MAX+1]; 30 struct Qdisc *queues[TCQ_PRIO_BANDS]; 31 }; 32 33 34 static struct Qdisc * 35 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr) 36 { 37 struct prio_sched_data *q = qdisc_priv(sch); 38 u32 band = skb->priority; 39 struct tcf_result res; 40 struct tcf_proto *fl; 41 int err; 42 43 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 44 if (TC_H_MAJ(skb->priority) != sch->handle) { 45 fl = rcu_dereference_bh(q->filter_list); 46 err = tcf_classify(skb, fl, &res, false); 47 #ifdef CONFIG_NET_CLS_ACT 48 switch (err) { 49 case TC_ACT_STOLEN: 50 case TC_ACT_QUEUED: 51 case TC_ACT_TRAP: 52 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 53 /* fall through */ 54 case TC_ACT_SHOT: 55 return NULL; 56 } 57 #endif 58 if (!fl || err < 0) { 59 if (TC_H_MAJ(band)) 60 band = 0; 61 return q->queues[q->prio2band[band & TC_PRIO_MAX]]; 62 } 63 band = res.classid; 64 } 65 band = TC_H_MIN(band) - 1; 66 if (band >= q->bands) 67 return q->queues[q->prio2band[0]]; 68 69 return q->queues[band]; 70 } 71 72 static int 73 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free) 74 { 75 struct Qdisc *qdisc; 76 int ret; 77 78 qdisc = prio_classify(skb, sch, &ret); 79 #ifdef CONFIG_NET_CLS_ACT 80 if (qdisc == NULL) { 81 82 if (ret & __NET_XMIT_BYPASS) 83 qdisc_qstats_drop(sch); 84 __qdisc_drop(skb, to_free); 85 return ret; 86 } 87 #endif 88 89 ret = qdisc_enqueue(skb, qdisc, to_free); 90 if (ret == NET_XMIT_SUCCESS) { 91 qdisc_qstats_backlog_inc(sch, skb); 92 sch->q.qlen++; 93 return NET_XMIT_SUCCESS; 94 } 95 if (net_xmit_drop_count(ret)) 96 qdisc_qstats_drop(sch); 97 return ret; 98 } 99 100 static struct sk_buff *prio_peek(struct Qdisc *sch) 101 { 102 struct prio_sched_data *q = qdisc_priv(sch); 103 int prio; 104 105 for (prio = 0; prio < q->bands; prio++) { 106 struct Qdisc *qdisc = q->queues[prio]; 107 struct sk_buff *skb = qdisc->ops->peek(qdisc); 108 if (skb) 109 return skb; 110 } 111 return NULL; 112 } 113 114 static struct sk_buff *prio_dequeue(struct Qdisc *sch) 115 { 116 struct prio_sched_data *q = qdisc_priv(sch); 117 int prio; 118 119 for (prio = 0; prio < q->bands; prio++) { 120 struct Qdisc *qdisc = q->queues[prio]; 121 struct sk_buff *skb = qdisc_dequeue_peeked(qdisc); 122 if (skb) { 123 qdisc_bstats_update(sch, skb); 124 qdisc_qstats_backlog_dec(sch, skb); 125 sch->q.qlen--; 126 return skb; 127 } 128 } 129 return NULL; 130 131 } 132 133 static void 134 prio_reset(struct Qdisc *sch) 135 { 136 int prio; 137 struct prio_sched_data *q = qdisc_priv(sch); 138 139 for (prio = 0; prio < q->bands; prio++) 140 qdisc_reset(q->queues[prio]); 141 sch->qstats.backlog = 0; 142 sch->q.qlen = 0; 143 } 144 145 static void 146 prio_destroy(struct Qdisc *sch) 147 { 148 int prio; 149 struct prio_sched_data *q = qdisc_priv(sch); 150 151 tcf_block_put(q->block); 152 for (prio = 0; prio < q->bands; prio++) 153 qdisc_destroy(q->queues[prio]); 154 } 155 156 static int prio_tune(struct Qdisc *sch, struct nlattr *opt) 157 { 158 struct prio_sched_data *q = qdisc_priv(sch); 159 struct Qdisc *queues[TCQ_PRIO_BANDS]; 160 int oldbands = q->bands, i; 161 struct tc_prio_qopt *qopt; 162 163 if (nla_len(opt) < sizeof(*qopt)) 164 return -EINVAL; 165 qopt = nla_data(opt); 166 167 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2) 168 return -EINVAL; 169 170 for (i = 0; i <= TC_PRIO_MAX; i++) { 171 if (qopt->priomap[i] >= qopt->bands) 172 return -EINVAL; 173 } 174 175 /* Before commit, make sure we can allocate all new qdiscs */ 176 for (i = oldbands; i < qopt->bands; i++) { 177 queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 178 TC_H_MAKE(sch->handle, i + 1)); 179 if (!queues[i]) { 180 while (i > oldbands) 181 qdisc_destroy(queues[--i]); 182 return -ENOMEM; 183 } 184 } 185 186 sch_tree_lock(sch); 187 q->bands = qopt->bands; 188 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1); 189 190 for (i = q->bands; i < oldbands; i++) { 191 struct Qdisc *child = q->queues[i]; 192 193 qdisc_tree_reduce_backlog(child, child->q.qlen, 194 child->qstats.backlog); 195 qdisc_destroy(child); 196 } 197 198 for (i = oldbands; i < q->bands; i++) { 199 q->queues[i] = queues[i]; 200 if (q->queues[i] != &noop_qdisc) 201 qdisc_hash_add(q->queues[i], true); 202 } 203 204 sch_tree_unlock(sch); 205 return 0; 206 } 207 208 static int prio_init(struct Qdisc *sch, struct nlattr *opt) 209 { 210 struct prio_sched_data *q = qdisc_priv(sch); 211 int err; 212 213 if (!opt) 214 return -EINVAL; 215 216 err = tcf_block_get(&q->block, &q->filter_list, sch); 217 if (err) 218 return err; 219 220 return prio_tune(sch, opt); 221 } 222 223 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb) 224 { 225 struct prio_sched_data *q = qdisc_priv(sch); 226 unsigned char *b = skb_tail_pointer(skb); 227 struct tc_prio_qopt opt; 228 229 opt.bands = q->bands; 230 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1); 231 232 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) 233 goto nla_put_failure; 234 235 return skb->len; 236 237 nla_put_failure: 238 nlmsg_trim(skb, b); 239 return -1; 240 } 241 242 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 243 struct Qdisc **old) 244 { 245 struct prio_sched_data *q = qdisc_priv(sch); 246 unsigned long band = arg - 1; 247 248 if (new == NULL) 249 new = &noop_qdisc; 250 251 *old = qdisc_replace(sch, new, &q->queues[band]); 252 return 0; 253 } 254 255 static struct Qdisc * 256 prio_leaf(struct Qdisc *sch, unsigned long arg) 257 { 258 struct prio_sched_data *q = qdisc_priv(sch); 259 unsigned long band = arg - 1; 260 261 return q->queues[band]; 262 } 263 264 static unsigned long prio_find(struct Qdisc *sch, u32 classid) 265 { 266 struct prio_sched_data *q = qdisc_priv(sch); 267 unsigned long band = TC_H_MIN(classid); 268 269 if (band - 1 >= q->bands) 270 return 0; 271 return band; 272 } 273 274 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid) 275 { 276 return prio_find(sch, classid); 277 } 278 279 280 static void prio_unbind(struct Qdisc *q, unsigned long cl) 281 { 282 } 283 284 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb, 285 struct tcmsg *tcm) 286 { 287 struct prio_sched_data *q = qdisc_priv(sch); 288 289 tcm->tcm_handle |= TC_H_MIN(cl); 290 tcm->tcm_info = q->queues[cl-1]->handle; 291 return 0; 292 } 293 294 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl, 295 struct gnet_dump *d) 296 { 297 struct prio_sched_data *q = qdisc_priv(sch); 298 struct Qdisc *cl_q; 299 300 cl_q = q->queues[cl - 1]; 301 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), 302 d, NULL, &cl_q->bstats) < 0 || 303 gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0) 304 return -1; 305 306 return 0; 307 } 308 309 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg) 310 { 311 struct prio_sched_data *q = qdisc_priv(sch); 312 int prio; 313 314 if (arg->stop) 315 return; 316 317 for (prio = 0; prio < q->bands; prio++) { 318 if (arg->count < arg->skip) { 319 arg->count++; 320 continue; 321 } 322 if (arg->fn(sch, prio + 1, arg) < 0) { 323 arg->stop = 1; 324 break; 325 } 326 arg->count++; 327 } 328 } 329 330 static struct tcf_block *prio_tcf_block(struct Qdisc *sch, unsigned long cl) 331 { 332 struct prio_sched_data *q = qdisc_priv(sch); 333 334 if (cl) 335 return NULL; 336 return q->block; 337 } 338 339 static const struct Qdisc_class_ops prio_class_ops = { 340 .graft = prio_graft, 341 .leaf = prio_leaf, 342 .find = prio_find, 343 .walk = prio_walk, 344 .tcf_block = prio_tcf_block, 345 .bind_tcf = prio_bind, 346 .unbind_tcf = prio_unbind, 347 .dump = prio_dump_class, 348 .dump_stats = prio_dump_class_stats, 349 }; 350 351 static struct Qdisc_ops prio_qdisc_ops __read_mostly = { 352 .next = NULL, 353 .cl_ops = &prio_class_ops, 354 .id = "prio", 355 .priv_size = sizeof(struct prio_sched_data), 356 .enqueue = prio_enqueue, 357 .dequeue = prio_dequeue, 358 .peek = prio_peek, 359 .init = prio_init, 360 .reset = prio_reset, 361 .destroy = prio_destroy, 362 .change = prio_tune, 363 .dump = prio_dump, 364 .owner = THIS_MODULE, 365 }; 366 367 static int __init prio_module_init(void) 368 { 369 return register_qdisc(&prio_qdisc_ops); 370 } 371 372 static void __exit prio_module_exit(void) 373 { 374 unregister_qdisc(&prio_qdisc_ops); 375 } 376 377 module_init(prio_module_init) 378 module_exit(prio_module_exit) 379 380 MODULE_LICENSE("GPL"); 381