1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * net/sched/sch_prio.c Simple 3-band priority "scheduler". 4 * 5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 6 * Fixes: 19990609: J Hadi Salim <hadi@nortelnetworks.com>: 7 * Init -- EINVAL when opt undefined 8 */ 9 10 #include <linux/module.h> 11 #include <linux/slab.h> 12 #include <linux/types.h> 13 #include <linux/kernel.h> 14 #include <linux/string.h> 15 #include <linux/errno.h> 16 #include <linux/skbuff.h> 17 #include <net/netlink.h> 18 #include <net/pkt_sched.h> 19 #include <net/pkt_cls.h> 20 21 struct prio_sched_data { 22 int bands; 23 struct tcf_proto __rcu *filter_list; 24 struct tcf_block *block; 25 u8 prio2band[TC_PRIO_MAX+1]; 26 struct Qdisc *queues[TCQ_PRIO_BANDS]; 27 }; 28 29 30 static struct Qdisc * 31 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr) 32 { 33 struct prio_sched_data *q = qdisc_priv(sch); 34 u32 band = skb->priority; 35 struct tcf_result res; 36 struct tcf_proto *fl; 37 int err; 38 39 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 40 if (TC_H_MAJ(skb->priority) != sch->handle) { 41 fl = rcu_dereference_bh(q->filter_list); 42 err = tcf_classify(skb, fl, &res, false); 43 #ifdef CONFIG_NET_CLS_ACT 44 switch (err) { 45 case TC_ACT_STOLEN: 46 case TC_ACT_QUEUED: 47 case TC_ACT_TRAP: 48 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 49 /* fall through */ 50 case TC_ACT_SHOT: 51 return NULL; 52 } 53 #endif 54 if (!fl || err < 0) { 55 if (TC_H_MAJ(band)) 56 band = 0; 57 return q->queues[q->prio2band[band & TC_PRIO_MAX]]; 58 } 59 band = res.classid; 60 } 61 band = TC_H_MIN(band) - 1; 62 if (band >= q->bands) 63 return q->queues[q->prio2band[0]]; 64 65 return q->queues[band]; 66 } 67 68 static int 69 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free) 70 { 71 unsigned int len = qdisc_pkt_len(skb); 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 __qdisc_drop(skb, to_free); 82 return ret; 83 } 84 #endif 85 86 ret = qdisc_enqueue(skb, qdisc, to_free); 87 if (ret == NET_XMIT_SUCCESS) { 88 sch->qstats.backlog += len; 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 void 131 prio_reset(struct Qdisc *sch) 132 { 133 int prio; 134 struct prio_sched_data *q = qdisc_priv(sch); 135 136 for (prio = 0; prio < q->bands; prio++) 137 qdisc_reset(q->queues[prio]); 138 sch->qstats.backlog = 0; 139 sch->q.qlen = 0; 140 } 141 142 static int prio_offload(struct Qdisc *sch, struct tc_prio_qopt *qopt) 143 { 144 struct net_device *dev = qdisc_dev(sch); 145 struct tc_prio_qopt_offload opt = { 146 .handle = sch->handle, 147 .parent = sch->parent, 148 }; 149 150 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 151 return -EOPNOTSUPP; 152 153 if (qopt) { 154 opt.command = TC_PRIO_REPLACE; 155 opt.replace_params.bands = qopt->bands; 156 memcpy(&opt.replace_params.priomap, qopt->priomap, 157 TC_PRIO_MAX + 1); 158 opt.replace_params.qstats = &sch->qstats; 159 } else { 160 opt.command = TC_PRIO_DESTROY; 161 } 162 163 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_PRIO, &opt); 164 } 165 166 static void 167 prio_destroy(struct Qdisc *sch) 168 { 169 int prio; 170 struct prio_sched_data *q = qdisc_priv(sch); 171 172 tcf_block_put(q->block); 173 prio_offload(sch, NULL); 174 for (prio = 0; prio < q->bands; prio++) 175 qdisc_put(q->queues[prio]); 176 } 177 178 static int prio_tune(struct Qdisc *sch, struct nlattr *opt, 179 struct netlink_ext_ack *extack) 180 { 181 struct prio_sched_data *q = qdisc_priv(sch); 182 struct Qdisc *queues[TCQ_PRIO_BANDS]; 183 int oldbands = q->bands, i; 184 struct tc_prio_qopt *qopt; 185 186 if (nla_len(opt) < sizeof(*qopt)) 187 return -EINVAL; 188 qopt = nla_data(opt); 189 190 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2) 191 return -EINVAL; 192 193 for (i = 0; i <= TC_PRIO_MAX; i++) { 194 if (qopt->priomap[i] >= qopt->bands) 195 return -EINVAL; 196 } 197 198 /* Before commit, make sure we can allocate all new qdiscs */ 199 for (i = oldbands; i < qopt->bands; i++) { 200 queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 201 TC_H_MAKE(sch->handle, i + 1), 202 extack); 203 if (!queues[i]) { 204 while (i > oldbands) 205 qdisc_put(queues[--i]); 206 return -ENOMEM; 207 } 208 } 209 210 prio_offload(sch, qopt); 211 sch_tree_lock(sch); 212 q->bands = qopt->bands; 213 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1); 214 215 for (i = q->bands; i < oldbands; i++) 216 qdisc_tree_flush_backlog(q->queues[i]); 217 218 for (i = oldbands; i < q->bands; i++) { 219 q->queues[i] = queues[i]; 220 if (q->queues[i] != &noop_qdisc) 221 qdisc_hash_add(q->queues[i], true); 222 } 223 224 sch_tree_unlock(sch); 225 226 for (i = q->bands; i < oldbands; i++) 227 qdisc_put(q->queues[i]); 228 return 0; 229 } 230 231 static int prio_init(struct Qdisc *sch, struct nlattr *opt, 232 struct netlink_ext_ack *extack) 233 { 234 struct prio_sched_data *q = qdisc_priv(sch); 235 int err; 236 237 if (!opt) 238 return -EINVAL; 239 240 err = tcf_block_get(&q->block, &q->filter_list, sch, extack); 241 if (err) 242 return err; 243 244 return prio_tune(sch, opt, extack); 245 } 246 247 static int prio_dump_offload(struct Qdisc *sch) 248 { 249 struct tc_prio_qopt_offload hw_stats = { 250 .command = TC_PRIO_STATS, 251 .handle = sch->handle, 252 .parent = sch->parent, 253 { 254 .stats = { 255 .bstats = &sch->bstats, 256 .qstats = &sch->qstats, 257 }, 258 }, 259 }; 260 261 return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_PRIO, &hw_stats); 262 } 263 264 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb) 265 { 266 struct prio_sched_data *q = qdisc_priv(sch); 267 unsigned char *b = skb_tail_pointer(skb); 268 struct tc_prio_qopt opt; 269 int err; 270 271 opt.bands = q->bands; 272 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1); 273 274 err = prio_dump_offload(sch); 275 if (err) 276 goto nla_put_failure; 277 278 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) 279 goto nla_put_failure; 280 281 return skb->len; 282 283 nla_put_failure: 284 nlmsg_trim(skb, b); 285 return -1; 286 } 287 288 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 289 struct Qdisc **old, struct netlink_ext_ack *extack) 290 { 291 struct prio_sched_data *q = qdisc_priv(sch); 292 struct tc_prio_qopt_offload graft_offload; 293 unsigned long band = arg - 1; 294 295 if (new == NULL) 296 new = &noop_qdisc; 297 298 *old = qdisc_replace(sch, new, &q->queues[band]); 299 300 graft_offload.handle = sch->handle; 301 graft_offload.parent = sch->parent; 302 graft_offload.graft_params.band = band; 303 graft_offload.graft_params.child_handle = new->handle; 304 graft_offload.command = TC_PRIO_GRAFT; 305 306 qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, *old, 307 TC_SETUP_QDISC_PRIO, &graft_offload, 308 extack); 309 return 0; 310 } 311 312 static struct Qdisc * 313 prio_leaf(struct Qdisc *sch, unsigned long arg) 314 { 315 struct prio_sched_data *q = qdisc_priv(sch); 316 unsigned long band = arg - 1; 317 318 return q->queues[band]; 319 } 320 321 static unsigned long prio_find(struct Qdisc *sch, u32 classid) 322 { 323 struct prio_sched_data *q = qdisc_priv(sch); 324 unsigned long band = TC_H_MIN(classid); 325 326 if (band - 1 >= q->bands) 327 return 0; 328 return band; 329 } 330 331 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid) 332 { 333 return prio_find(sch, classid); 334 } 335 336 337 static void prio_unbind(struct Qdisc *q, unsigned long cl) 338 { 339 } 340 341 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb, 342 struct tcmsg *tcm) 343 { 344 struct prio_sched_data *q = qdisc_priv(sch); 345 346 tcm->tcm_handle |= TC_H_MIN(cl); 347 tcm->tcm_info = q->queues[cl-1]->handle; 348 return 0; 349 } 350 351 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl, 352 struct gnet_dump *d) 353 { 354 struct prio_sched_data *q = qdisc_priv(sch); 355 struct Qdisc *cl_q; 356 357 cl_q = q->queues[cl - 1]; 358 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), 359 d, cl_q->cpu_bstats, &cl_q->bstats) < 0 || 360 qdisc_qstats_copy(d, cl_q) < 0) 361 return -1; 362 363 return 0; 364 } 365 366 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg) 367 { 368 struct prio_sched_data *q = qdisc_priv(sch); 369 int prio; 370 371 if (arg->stop) 372 return; 373 374 for (prio = 0; prio < q->bands; prio++) { 375 if (arg->count < arg->skip) { 376 arg->count++; 377 continue; 378 } 379 if (arg->fn(sch, prio + 1, arg) < 0) { 380 arg->stop = 1; 381 break; 382 } 383 arg->count++; 384 } 385 } 386 387 static struct tcf_block *prio_tcf_block(struct Qdisc *sch, unsigned long cl, 388 struct netlink_ext_ack *extack) 389 { 390 struct prio_sched_data *q = qdisc_priv(sch); 391 392 if (cl) 393 return NULL; 394 return q->block; 395 } 396 397 static const struct Qdisc_class_ops prio_class_ops = { 398 .graft = prio_graft, 399 .leaf = prio_leaf, 400 .find = prio_find, 401 .walk = prio_walk, 402 .tcf_block = prio_tcf_block, 403 .bind_tcf = prio_bind, 404 .unbind_tcf = prio_unbind, 405 .dump = prio_dump_class, 406 .dump_stats = prio_dump_class_stats, 407 }; 408 409 static struct Qdisc_ops prio_qdisc_ops __read_mostly = { 410 .next = NULL, 411 .cl_ops = &prio_class_ops, 412 .id = "prio", 413 .priv_size = sizeof(struct prio_sched_data), 414 .enqueue = prio_enqueue, 415 .dequeue = prio_dequeue, 416 .peek = prio_peek, 417 .init = prio_init, 418 .reset = prio_reset, 419 .destroy = prio_destroy, 420 .change = prio_tune, 421 .dump = prio_dump, 422 .owner = THIS_MODULE, 423 }; 424 425 static int __init prio_module_init(void) 426 { 427 return register_qdisc(&prio_qdisc_ops); 428 } 429 430 static void __exit prio_module_exit(void) 431 { 432 unregister_qdisc(&prio_qdisc_ops); 433 } 434 435 module_init(prio_module_init) 436 module_exit(prio_module_exit) 437 438 MODULE_LICENSE("GPL"); 439