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, NULL, 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 fallthrough; 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 } 139 140 static int prio_offload(struct Qdisc *sch, struct tc_prio_qopt *qopt) 141 { 142 struct net_device *dev = qdisc_dev(sch); 143 struct tc_prio_qopt_offload opt = { 144 .handle = sch->handle, 145 .parent = sch->parent, 146 }; 147 148 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 149 return -EOPNOTSUPP; 150 151 if (qopt) { 152 opt.command = TC_PRIO_REPLACE; 153 opt.replace_params.bands = qopt->bands; 154 memcpy(&opt.replace_params.priomap, qopt->priomap, 155 TC_PRIO_MAX + 1); 156 opt.replace_params.qstats = &sch->qstats; 157 } else { 158 opt.command = TC_PRIO_DESTROY; 159 } 160 161 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_PRIO, &opt); 162 } 163 164 static void 165 prio_destroy(struct Qdisc *sch) 166 { 167 int prio; 168 struct prio_sched_data *q = qdisc_priv(sch); 169 170 tcf_block_put(q->block); 171 prio_offload(sch, NULL); 172 for (prio = 0; prio < q->bands; prio++) 173 qdisc_put(q->queues[prio]); 174 } 175 176 static int prio_tune(struct Qdisc *sch, struct nlattr *opt, 177 struct netlink_ext_ack *extack) 178 { 179 struct prio_sched_data *q = qdisc_priv(sch); 180 struct Qdisc *queues[TCQ_PRIO_BANDS]; 181 int oldbands = q->bands, i; 182 struct tc_prio_qopt *qopt; 183 184 if (nla_len(opt) < sizeof(*qopt)) 185 return -EINVAL; 186 qopt = nla_data(opt); 187 188 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < TCQ_MIN_PRIO_BANDS) 189 return -EINVAL; 190 191 for (i = 0; i <= TC_PRIO_MAX; i++) { 192 if (qopt->priomap[i] >= qopt->bands) 193 return -EINVAL; 194 } 195 196 /* Before commit, make sure we can allocate all new qdiscs */ 197 for (i = oldbands; i < qopt->bands; i++) { 198 queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 199 TC_H_MAKE(sch->handle, i + 1), 200 extack); 201 if (!queues[i]) { 202 while (i > oldbands) 203 qdisc_put(queues[--i]); 204 return -ENOMEM; 205 } 206 } 207 208 prio_offload(sch, qopt); 209 sch_tree_lock(sch); 210 q->bands = qopt->bands; 211 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1); 212 213 for (i = q->bands; i < oldbands; i++) 214 qdisc_tree_flush_backlog(q->queues[i]); 215 216 for (i = oldbands; i < q->bands; i++) { 217 q->queues[i] = queues[i]; 218 if (q->queues[i] != &noop_qdisc) 219 qdisc_hash_add(q->queues[i], true); 220 } 221 222 sch_tree_unlock(sch); 223 224 for (i = q->bands; i < oldbands; i++) 225 qdisc_put(q->queues[i]); 226 return 0; 227 } 228 229 static int prio_init(struct Qdisc *sch, struct nlattr *opt, 230 struct netlink_ext_ack *extack) 231 { 232 struct prio_sched_data *q = qdisc_priv(sch); 233 int err; 234 235 if (!opt) 236 return -EINVAL; 237 238 err = tcf_block_get(&q->block, &q->filter_list, sch, extack); 239 if (err) 240 return err; 241 242 return prio_tune(sch, opt, extack); 243 } 244 245 static int prio_dump_offload(struct Qdisc *sch) 246 { 247 struct tc_prio_qopt_offload hw_stats = { 248 .command = TC_PRIO_STATS, 249 .handle = sch->handle, 250 .parent = sch->parent, 251 { 252 .stats = { 253 .bstats = &sch->bstats, 254 .qstats = &sch->qstats, 255 }, 256 }, 257 }; 258 259 return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_PRIO, &hw_stats); 260 } 261 262 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb) 263 { 264 struct prio_sched_data *q = qdisc_priv(sch); 265 unsigned char *b = skb_tail_pointer(skb); 266 struct tc_prio_qopt opt; 267 int err; 268 269 opt.bands = q->bands; 270 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1); 271 272 err = prio_dump_offload(sch); 273 if (err) 274 goto nla_put_failure; 275 276 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) 277 goto nla_put_failure; 278 279 return skb->len; 280 281 nla_put_failure: 282 nlmsg_trim(skb, b); 283 return -1; 284 } 285 286 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 287 struct Qdisc **old, struct netlink_ext_ack *extack) 288 { 289 struct prio_sched_data *q = qdisc_priv(sch); 290 struct tc_prio_qopt_offload graft_offload; 291 unsigned long band = arg - 1; 292 293 if (!new) { 294 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 295 TC_H_MAKE(sch->handle, arg), extack); 296 if (!new) 297 new = &noop_qdisc; 298 else 299 qdisc_hash_add(new, true); 300 } 301 302 *old = qdisc_replace(sch, new, &q->queues[band]); 303 304 graft_offload.handle = sch->handle; 305 graft_offload.parent = sch->parent; 306 graft_offload.graft_params.band = band; 307 graft_offload.graft_params.child_handle = new->handle; 308 graft_offload.command = TC_PRIO_GRAFT; 309 310 qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, *old, 311 TC_SETUP_QDISC_PRIO, &graft_offload, 312 extack); 313 return 0; 314 } 315 316 static struct Qdisc * 317 prio_leaf(struct Qdisc *sch, unsigned long arg) 318 { 319 struct prio_sched_data *q = qdisc_priv(sch); 320 unsigned long band = arg - 1; 321 322 return q->queues[band]; 323 } 324 325 static unsigned long prio_find(struct Qdisc *sch, u32 classid) 326 { 327 struct prio_sched_data *q = qdisc_priv(sch); 328 unsigned long band = TC_H_MIN(classid); 329 330 if (band - 1 >= q->bands) 331 return 0; 332 return band; 333 } 334 335 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid) 336 { 337 return prio_find(sch, classid); 338 } 339 340 341 static void prio_unbind(struct Qdisc *q, unsigned long cl) 342 { 343 } 344 345 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb, 346 struct tcmsg *tcm) 347 { 348 struct prio_sched_data *q = qdisc_priv(sch); 349 350 tcm->tcm_handle |= TC_H_MIN(cl); 351 tcm->tcm_info = q->queues[cl-1]->handle; 352 return 0; 353 } 354 355 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl, 356 struct gnet_dump *d) 357 { 358 struct prio_sched_data *q = qdisc_priv(sch); 359 struct Qdisc *cl_q; 360 361 cl_q = q->queues[cl - 1]; 362 if (gnet_stats_copy_basic(d, cl_q->cpu_bstats, 363 &cl_q->bstats, true) < 0 || 364 qdisc_qstats_copy(d, cl_q) < 0) 365 return -1; 366 367 return 0; 368 } 369 370 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg) 371 { 372 struct prio_sched_data *q = qdisc_priv(sch); 373 int prio; 374 375 if (arg->stop) 376 return; 377 378 for (prio = 0; prio < q->bands; prio++) { 379 if (arg->count < arg->skip) { 380 arg->count++; 381 continue; 382 } 383 if (arg->fn(sch, prio + 1, arg) < 0) { 384 arg->stop = 1; 385 break; 386 } 387 arg->count++; 388 } 389 } 390 391 static struct tcf_block *prio_tcf_block(struct Qdisc *sch, unsigned long cl, 392 struct netlink_ext_ack *extack) 393 { 394 struct prio_sched_data *q = qdisc_priv(sch); 395 396 if (cl) 397 return NULL; 398 return q->block; 399 } 400 401 static const struct Qdisc_class_ops prio_class_ops = { 402 .graft = prio_graft, 403 .leaf = prio_leaf, 404 .find = prio_find, 405 .walk = prio_walk, 406 .tcf_block = prio_tcf_block, 407 .bind_tcf = prio_bind, 408 .unbind_tcf = prio_unbind, 409 .dump = prio_dump_class, 410 .dump_stats = prio_dump_class_stats, 411 }; 412 413 static struct Qdisc_ops prio_qdisc_ops __read_mostly = { 414 .next = NULL, 415 .cl_ops = &prio_class_ops, 416 .id = "prio", 417 .priv_size = sizeof(struct prio_sched_data), 418 .enqueue = prio_enqueue, 419 .dequeue = prio_dequeue, 420 .peek = prio_peek, 421 .init = prio_init, 422 .reset = prio_reset, 423 .destroy = prio_destroy, 424 .change = prio_tune, 425 .dump = prio_dump, 426 .owner = THIS_MODULE, 427 }; 428 429 static int __init prio_module_init(void) 430 { 431 return register_qdisc(&prio_qdisc_ops); 432 } 433 434 static void __exit prio_module_exit(void) 435 { 436 unregister_qdisc(&prio_qdisc_ops); 437 } 438 439 module_init(prio_module_init) 440 module_exit(prio_module_exit) 441 442 MODULE_LICENSE("GPL"); 443