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 24 25 struct prio_sched_data { 26 int bands; 27 struct tcf_proto __rcu *filter_list; 28 u8 prio2band[TC_PRIO_MAX+1]; 29 struct Qdisc *queues[TCQ_PRIO_BANDS]; 30 }; 31 32 33 static struct Qdisc * 34 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr) 35 { 36 struct prio_sched_data *q = qdisc_priv(sch); 37 u32 band = skb->priority; 38 struct tcf_result res; 39 struct tcf_proto *fl; 40 int err; 41 42 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 43 if (TC_H_MAJ(skb->priority) != sch->handle) { 44 fl = rcu_dereference_bh(q->filter_list); 45 err = tc_classify(skb, fl, &res, false); 46 #ifdef CONFIG_NET_CLS_ACT 47 switch (err) { 48 case TC_ACT_STOLEN: 49 case TC_ACT_QUEUED: 50 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 51 case TC_ACT_SHOT: 52 return NULL; 53 } 54 #endif 55 if (!fl || err < 0) { 56 if (TC_H_MAJ(band)) 57 band = 0; 58 return q->queues[q->prio2band[band & TC_PRIO_MAX]]; 59 } 60 band = res.classid; 61 } 62 band = TC_H_MIN(band) - 1; 63 if (band >= q->bands) 64 return q->queues[q->prio2band[0]]; 65 66 return q->queues[band]; 67 } 68 69 static int 70 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch) 71 { 72 struct Qdisc *qdisc; 73 int ret; 74 75 qdisc = prio_classify(skb, sch, &ret); 76 #ifdef CONFIG_NET_CLS_ACT 77 if (qdisc == NULL) { 78 79 if (ret & __NET_XMIT_BYPASS) 80 qdisc_qstats_drop(sch); 81 kfree_skb(skb); 82 return ret; 83 } 84 #endif 85 86 ret = qdisc_enqueue(skb, qdisc); 87 if (ret == NET_XMIT_SUCCESS) { 88 qdisc_qstats_backlog_inc(sch, skb); 89 sch->q.qlen++; 90 return NET_XMIT_SUCCESS; 91 } 92 if (net_xmit_drop_count(ret)) 93 qdisc_qstats_drop(sch); 94 return ret; 95 } 96 97 static struct sk_buff *prio_peek(struct Qdisc *sch) 98 { 99 struct prio_sched_data *q = qdisc_priv(sch); 100 int prio; 101 102 for (prio = 0; prio < q->bands; prio++) { 103 struct Qdisc *qdisc = q->queues[prio]; 104 struct sk_buff *skb = qdisc->ops->peek(qdisc); 105 if (skb) 106 return skb; 107 } 108 return NULL; 109 } 110 111 static struct sk_buff *prio_dequeue(struct Qdisc *sch) 112 { 113 struct prio_sched_data *q = qdisc_priv(sch); 114 int prio; 115 116 for (prio = 0; prio < q->bands; prio++) { 117 struct Qdisc *qdisc = q->queues[prio]; 118 struct sk_buff *skb = qdisc_dequeue_peeked(qdisc); 119 if (skb) { 120 qdisc_bstats_update(sch, skb); 121 qdisc_qstats_backlog_dec(sch, skb); 122 sch->q.qlen--; 123 return skb; 124 } 125 } 126 return NULL; 127 128 } 129 130 static unsigned int prio_drop(struct Qdisc *sch) 131 { 132 struct prio_sched_data *q = qdisc_priv(sch); 133 int prio; 134 unsigned int len; 135 struct Qdisc *qdisc; 136 137 for (prio = q->bands-1; prio >= 0; prio--) { 138 qdisc = q->queues[prio]; 139 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) { 140 sch->qstats.backlog -= len; 141 sch->q.qlen--; 142 return len; 143 } 144 } 145 return 0; 146 } 147 148 149 static void 150 prio_reset(struct Qdisc *sch) 151 { 152 int prio; 153 struct prio_sched_data *q = qdisc_priv(sch); 154 155 for (prio = 0; prio < q->bands; prio++) 156 qdisc_reset(q->queues[prio]); 157 sch->qstats.backlog = 0; 158 sch->q.qlen = 0; 159 } 160 161 static void 162 prio_destroy(struct Qdisc *sch) 163 { 164 int prio; 165 struct prio_sched_data *q = qdisc_priv(sch); 166 167 tcf_destroy_chain(&q->filter_list); 168 for (prio = 0; prio < q->bands; prio++) 169 qdisc_destroy(q->queues[prio]); 170 } 171 172 static int prio_tune(struct Qdisc *sch, struct nlattr *opt) 173 { 174 struct prio_sched_data *q = qdisc_priv(sch); 175 struct tc_prio_qopt *qopt; 176 int i; 177 178 if (nla_len(opt) < sizeof(*qopt)) 179 return -EINVAL; 180 qopt = nla_data(opt); 181 182 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2) 183 return -EINVAL; 184 185 for (i = 0; i <= TC_PRIO_MAX; i++) { 186 if (qopt->priomap[i] >= qopt->bands) 187 return -EINVAL; 188 } 189 190 sch_tree_lock(sch); 191 q->bands = qopt->bands; 192 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1); 193 194 for (i = q->bands; i < TCQ_PRIO_BANDS; i++) { 195 struct Qdisc *child = q->queues[i]; 196 q->queues[i] = &noop_qdisc; 197 if (child != &noop_qdisc) { 198 qdisc_tree_reduce_backlog(child, child->q.qlen, child->qstats.backlog); 199 qdisc_destroy(child); 200 } 201 } 202 sch_tree_unlock(sch); 203 204 for (i = 0; i < q->bands; i++) { 205 if (q->queues[i] == &noop_qdisc) { 206 struct Qdisc *child, *old; 207 208 child = qdisc_create_dflt(sch->dev_queue, 209 &pfifo_qdisc_ops, 210 TC_H_MAKE(sch->handle, i + 1)); 211 if (child) { 212 sch_tree_lock(sch); 213 old = q->queues[i]; 214 q->queues[i] = child; 215 216 if (old != &noop_qdisc) { 217 qdisc_tree_reduce_backlog(old, 218 old->q.qlen, 219 old->qstats.backlog); 220 qdisc_destroy(old); 221 } 222 sch_tree_unlock(sch); 223 } 224 } 225 } 226 return 0; 227 } 228 229 static int prio_init(struct Qdisc *sch, struct nlattr *opt) 230 { 231 struct prio_sched_data *q = qdisc_priv(sch); 232 int i; 233 234 for (i = 0; i < TCQ_PRIO_BANDS; i++) 235 q->queues[i] = &noop_qdisc; 236 237 if (opt == NULL) { 238 return -EINVAL; 239 } else { 240 int err; 241 242 if ((err = prio_tune(sch, opt)) != 0) 243 return err; 244 } 245 return 0; 246 } 247 248 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb) 249 { 250 struct prio_sched_data *q = qdisc_priv(sch); 251 unsigned char *b = skb_tail_pointer(skb); 252 struct tc_prio_qopt opt; 253 254 opt.bands = q->bands; 255 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1); 256 257 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) 258 goto nla_put_failure; 259 260 return skb->len; 261 262 nla_put_failure: 263 nlmsg_trim(skb, b); 264 return -1; 265 } 266 267 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 268 struct Qdisc **old) 269 { 270 struct prio_sched_data *q = qdisc_priv(sch); 271 unsigned long band = arg - 1; 272 273 if (new == NULL) 274 new = &noop_qdisc; 275 276 *old = qdisc_replace(sch, new, &q->queues[band]); 277 return 0; 278 } 279 280 static struct Qdisc * 281 prio_leaf(struct Qdisc *sch, unsigned long arg) 282 { 283 struct prio_sched_data *q = qdisc_priv(sch); 284 unsigned long band = arg - 1; 285 286 return q->queues[band]; 287 } 288 289 static unsigned long prio_get(struct Qdisc *sch, u32 classid) 290 { 291 struct prio_sched_data *q = qdisc_priv(sch); 292 unsigned long band = TC_H_MIN(classid); 293 294 if (band - 1 >= q->bands) 295 return 0; 296 return band; 297 } 298 299 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid) 300 { 301 return prio_get(sch, classid); 302 } 303 304 305 static void prio_put(struct Qdisc *q, unsigned long cl) 306 { 307 } 308 309 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb, 310 struct tcmsg *tcm) 311 { 312 struct prio_sched_data *q = qdisc_priv(sch); 313 314 tcm->tcm_handle |= TC_H_MIN(cl); 315 tcm->tcm_info = q->queues[cl-1]->handle; 316 return 0; 317 } 318 319 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl, 320 struct gnet_dump *d) 321 { 322 struct prio_sched_data *q = qdisc_priv(sch); 323 struct Qdisc *cl_q; 324 325 cl_q = q->queues[cl - 1]; 326 if (gnet_stats_copy_basic(d, NULL, &cl_q->bstats) < 0 || 327 gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0) 328 return -1; 329 330 return 0; 331 } 332 333 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg) 334 { 335 struct prio_sched_data *q = qdisc_priv(sch); 336 int prio; 337 338 if (arg->stop) 339 return; 340 341 for (prio = 0; prio < q->bands; prio++) { 342 if (arg->count < arg->skip) { 343 arg->count++; 344 continue; 345 } 346 if (arg->fn(sch, prio + 1, arg) < 0) { 347 arg->stop = 1; 348 break; 349 } 350 arg->count++; 351 } 352 } 353 354 static struct tcf_proto __rcu **prio_find_tcf(struct Qdisc *sch, 355 unsigned long cl) 356 { 357 struct prio_sched_data *q = qdisc_priv(sch); 358 359 if (cl) 360 return NULL; 361 return &q->filter_list; 362 } 363 364 static const struct Qdisc_class_ops prio_class_ops = { 365 .graft = prio_graft, 366 .leaf = prio_leaf, 367 .get = prio_get, 368 .put = prio_put, 369 .walk = prio_walk, 370 .tcf_chain = prio_find_tcf, 371 .bind_tcf = prio_bind, 372 .unbind_tcf = prio_put, 373 .dump = prio_dump_class, 374 .dump_stats = prio_dump_class_stats, 375 }; 376 377 static struct Qdisc_ops prio_qdisc_ops __read_mostly = { 378 .next = NULL, 379 .cl_ops = &prio_class_ops, 380 .id = "prio", 381 .priv_size = sizeof(struct prio_sched_data), 382 .enqueue = prio_enqueue, 383 .dequeue = prio_dequeue, 384 .peek = prio_peek, 385 .drop = prio_drop, 386 .init = prio_init, 387 .reset = prio_reset, 388 .destroy = prio_destroy, 389 .change = prio_tune, 390 .dump = prio_dump, 391 .owner = THIS_MODULE, 392 }; 393 394 static int __init prio_module_init(void) 395 { 396 return register_qdisc(&prio_qdisc_ops); 397 } 398 399 static void __exit prio_module_exit(void) 400 { 401 unregister_qdisc(&prio_qdisc_ops); 402 } 403 404 module_init(prio_module_init) 405 module_exit(prio_module_exit) 406 407 MODULE_LICENSE("GPL"); 408