| 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 int prio_offload(struct Qdisc *sch, struct tc_prio_qopt *qopt) |
| 146 | { |
| 147 | struct net_device *dev = qdisc_dev(sch); |
| 148 | struct tc_prio_qopt_offload opt = { |
| 149 | .handle = sch->handle, |
| 150 | .parent = sch->parent, |
| 151 | }; |
| 152 | |
| 153 | if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) |
| 154 | return -EOPNOTSUPP; |
| 155 | |
| 156 | if (qopt) { |
| 157 | opt.command = TC_PRIO_REPLACE; |
| 158 | opt.replace_params.bands = qopt->bands; |
| 159 | memcpy(&opt.replace_params.priomap, qopt->priomap, |
| 160 | TC_PRIO_MAX + 1); |
| 161 | opt.replace_params.qstats = &sch->qstats; |
| 162 | } else { |
| 163 | opt.command = TC_PRIO_DESTROY; |
| 164 | } |
| 165 | |
| 166 | return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_PRIO, &opt); |
| 167 | } |
| 168 | |
| 169 | static void |
| 170 | prio_destroy(struct Qdisc *sch) |
| 171 | { |
| 172 | int prio; |
| 173 | struct prio_sched_data *q = qdisc_priv(sch); |
| 174 | |
| 175 | tcf_block_put(q->block); |
| 176 | prio_offload(sch, NULL); |
| 177 | for (prio = 0; prio < q->bands; prio++) |
| 178 | qdisc_destroy(q->queues[prio]); |
| 179 | } |
| 180 | |
| 181 | static int prio_tune(struct Qdisc *sch, struct nlattr *opt, |
| 182 | struct netlink_ext_ack *extack) |
| 183 | { |
| 184 | struct prio_sched_data *q = qdisc_priv(sch); |
| 185 | struct Qdisc *queues[TCQ_PRIO_BANDS]; |
| 186 | int oldbands = q->bands, i; |
| 187 | struct tc_prio_qopt *qopt; |
| 188 | |
| 189 | if (nla_len(opt) < sizeof(*qopt)) |
| 190 | return -EINVAL; |
| 191 | qopt = nla_data(opt); |
| 192 | |
| 193 | if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2) |
| 194 | return -EINVAL; |
| 195 | |
| 196 | for (i = 0; i <= TC_PRIO_MAX; i++) { |
| 197 | if (qopt->priomap[i] >= qopt->bands) |
| 198 | return -EINVAL; |
| 199 | } |
| 200 | |
| 201 | /* Before commit, make sure we can allocate all new qdiscs */ |
| 202 | for (i = oldbands; i < qopt->bands; i++) { |
| 203 | queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
| 204 | TC_H_MAKE(sch->handle, i + 1), |
| 205 | extack); |
| 206 | if (!queues[i]) { |
| 207 | while (i > oldbands) |
| 208 | qdisc_destroy(queues[--i]); |
| 209 | return -ENOMEM; |
| 210 | } |
| 211 | } |
| 212 | |
| 213 | prio_offload(sch, qopt); |
| 214 | sch_tree_lock(sch); |
| 215 | q->bands = qopt->bands; |
| 216 | memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1); |
| 217 | |
| 218 | for (i = q->bands; i < oldbands; i++) { |
| 219 | struct Qdisc *child = q->queues[i]; |
| 220 | |
| 221 | qdisc_tree_reduce_backlog(child, child->q.qlen, |
| 222 | child->qstats.backlog); |
| 223 | qdisc_destroy(child); |
| 224 | } |
| 225 | |
| 226 | for (i = oldbands; i < q->bands; i++) { |
| 227 | q->queues[i] = queues[i]; |
| 228 | if (q->queues[i] != &noop_qdisc) |
| 229 | qdisc_hash_add(q->queues[i], true); |
| 230 | } |
| 231 | |
| 232 | sch_tree_unlock(sch); |
| 233 | return 0; |
| 234 | } |
| 235 | |
| 236 | static int prio_init(struct Qdisc *sch, struct nlattr *opt, |
| 237 | struct netlink_ext_ack *extack) |
| 238 | { |
| 239 | struct prio_sched_data *q = qdisc_priv(sch); |
| 240 | int err; |
| 241 | |
| 242 | if (!opt) |
| 243 | return -EINVAL; |
| 244 | |
| 245 | err = tcf_block_get(&q->block, &q->filter_list, sch, extack); |
| 246 | if (err) |
| 247 | return err; |
| 248 | |
| 249 | return prio_tune(sch, opt, extack); |
| 250 | } |
| 251 | |
| 252 | static int prio_dump_offload(struct Qdisc *sch) |
| 253 | { |
| 254 | struct net_device *dev = qdisc_dev(sch); |
| 255 | struct tc_prio_qopt_offload hw_stats = { |
| 256 | .command = TC_PRIO_STATS, |
| 257 | .handle = sch->handle, |
| 258 | .parent = sch->parent, |
| 259 | { |
| 260 | .stats = { |
| 261 | .bstats = &sch->bstats, |
| 262 | .qstats = &sch->qstats, |
| 263 | }, |
| 264 | }, |
| 265 | }; |
| 266 | int err; |
| 267 | |
| 268 | sch->flags &= ~TCQ_F_OFFLOADED; |
| 269 | if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) |
| 270 | return 0; |
| 271 | |
| 272 | err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_PRIO, |
| 273 | &hw_stats); |
| 274 | if (err == -EOPNOTSUPP) |
| 275 | return 0; |
| 276 | |
| 277 | if (!err) |
| 278 | sch->flags |= TCQ_F_OFFLOADED; |
| 279 | |
| 280 | return err; |
| 281 | } |
| 282 | |
| 283 | static int prio_dump(struct Qdisc *sch, struct sk_buff *skb) |
| 284 | { |
| 285 | struct prio_sched_data *q = qdisc_priv(sch); |
| 286 | unsigned char *b = skb_tail_pointer(skb); |
| 287 | struct tc_prio_qopt opt; |
| 288 | int err; |
| 289 | |
| 290 | opt.bands = q->bands; |
| 291 | memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1); |
| 292 | |
| 293 | err = prio_dump_offload(sch); |
| 294 | if (err) |
| 295 | goto nla_put_failure; |
| 296 | |
| 297 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
| 298 | goto nla_put_failure; |
| 299 | |
| 300 | return skb->len; |
| 301 | |
| 302 | nla_put_failure: |
| 303 | nlmsg_trim(skb, b); |
| 304 | return -1; |
| 305 | } |
| 306 | |
| 307 | static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, |
| 308 | struct Qdisc **old, struct netlink_ext_ack *extack) |
| 309 | { |
| 310 | struct prio_sched_data *q = qdisc_priv(sch); |
| 311 | struct tc_prio_qopt_offload graft_offload; |
| 312 | struct net_device *dev = qdisc_dev(sch); |
| 313 | unsigned long band = arg - 1; |
| 314 | bool any_qdisc_is_offloaded; |
| 315 | int err; |
| 316 | |
| 317 | if (new == NULL) |
| 318 | new = &noop_qdisc; |
| 319 | |
| 320 | *old = qdisc_replace(sch, new, &q->queues[band]); |
| 321 | |
| 322 | if (!tc_can_offload(dev)) |
| 323 | return 0; |
| 324 | |
| 325 | graft_offload.handle = sch->handle; |
| 326 | graft_offload.parent = sch->parent; |
| 327 | graft_offload.graft_params.band = band; |
| 328 | graft_offload.graft_params.child_handle = new->handle; |
| 329 | graft_offload.command = TC_PRIO_GRAFT; |
| 330 | |
| 331 | err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_PRIO, |
| 332 | &graft_offload); |
| 333 | |
| 334 | /* Don't report error if the graft is part of destroy operation. */ |
| 335 | if (err && new != &noop_qdisc) { |
| 336 | /* Don't report error if the parent, the old child and the new |
| 337 | * one are not offloaded. |
| 338 | */ |
| 339 | any_qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED; |
| 340 | any_qdisc_is_offloaded |= new->flags & TCQ_F_OFFLOADED; |
| 341 | if (*old) |
| 342 | any_qdisc_is_offloaded |= (*old)->flags & |
| 343 | TCQ_F_OFFLOADED; |
| 344 | |
| 345 | if (any_qdisc_is_offloaded) |
| 346 | NL_SET_ERR_MSG(extack, "Offloading graft operation failed."); |
| 347 | } |
| 348 | |
| 349 | return 0; |
| 350 | } |
| 351 | |
| 352 | static struct Qdisc * |
| 353 | prio_leaf(struct Qdisc *sch, unsigned long arg) |
| 354 | { |
| 355 | struct prio_sched_data *q = qdisc_priv(sch); |
| 356 | unsigned long band = arg - 1; |
| 357 | |
| 358 | return q->queues[band]; |
| 359 | } |
| 360 | |
| 361 | static unsigned long prio_find(struct Qdisc *sch, u32 classid) |
| 362 | { |
| 363 | struct prio_sched_data *q = qdisc_priv(sch); |
| 364 | unsigned long band = TC_H_MIN(classid); |
| 365 | |
| 366 | if (band - 1 >= q->bands) |
| 367 | return 0; |
| 368 | return band; |
| 369 | } |
| 370 | |
| 371 | static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid) |
| 372 | { |
| 373 | return prio_find(sch, classid); |
| 374 | } |
| 375 | |
| 376 | |
| 377 | static void prio_unbind(struct Qdisc *q, unsigned long cl) |
| 378 | { |
| 379 | } |
| 380 | |
| 381 | static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb, |
| 382 | struct tcmsg *tcm) |
| 383 | { |
| 384 | struct prio_sched_data *q = qdisc_priv(sch); |
| 385 | |
| 386 | tcm->tcm_handle |= TC_H_MIN(cl); |
| 387 | tcm->tcm_info = q->queues[cl-1]->handle; |
| 388 | return 0; |
| 389 | } |
| 390 | |
| 391 | static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl, |
| 392 | struct gnet_dump *d) |
| 393 | { |
| 394 | struct prio_sched_data *q = qdisc_priv(sch); |
| 395 | struct Qdisc *cl_q; |
| 396 | |
| 397 | cl_q = q->queues[cl - 1]; |
| 398 | if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), |
| 399 | d, NULL, &cl_q->bstats) < 0 || |
| 400 | gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0) |
| 401 | return -1; |
| 402 | |
| 403 | return 0; |
| 404 | } |
| 405 | |
| 406 | static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg) |
| 407 | { |
| 408 | struct prio_sched_data *q = qdisc_priv(sch); |
| 409 | int prio; |
| 410 | |
| 411 | if (arg->stop) |
| 412 | return; |
| 413 | |
| 414 | for (prio = 0; prio < q->bands; prio++) { |
| 415 | if (arg->count < arg->skip) { |
| 416 | arg->count++; |
| 417 | continue; |
| 418 | } |
| 419 | if (arg->fn(sch, prio + 1, arg) < 0) { |
| 420 | arg->stop = 1; |
| 421 | break; |
| 422 | } |
| 423 | arg->count++; |
| 424 | } |
| 425 | } |
| 426 | |
| 427 | static struct tcf_block *prio_tcf_block(struct Qdisc *sch, unsigned long cl, |
| 428 | struct netlink_ext_ack *extack) |
| 429 | { |
| 430 | struct prio_sched_data *q = qdisc_priv(sch); |
| 431 | |
| 432 | if (cl) |
| 433 | return NULL; |
| 434 | return q->block; |
| 435 | } |
| 436 | |
| 437 | static const struct Qdisc_class_ops prio_class_ops = { |
| 438 | .graft = prio_graft, |
| 439 | .leaf = prio_leaf, |
| 440 | .find = prio_find, |
| 441 | .walk = prio_walk, |
| 442 | .tcf_block = prio_tcf_block, |
| 443 | .bind_tcf = prio_bind, |
| 444 | .unbind_tcf = prio_unbind, |
| 445 | .dump = prio_dump_class, |
| 446 | .dump_stats = prio_dump_class_stats, |
| 447 | }; |
| 448 | |
| 449 | static struct Qdisc_ops prio_qdisc_ops __read_mostly = { |
| 450 | .next = NULL, |
| 451 | .cl_ops = &prio_class_ops, |
| 452 | .id = "prio", |
| 453 | .priv_size = sizeof(struct prio_sched_data), |
| 454 | .enqueue = prio_enqueue, |
| 455 | .dequeue = prio_dequeue, |
| 456 | .peek = prio_peek, |
| 457 | .init = prio_init, |
| 458 | .reset = prio_reset, |
| 459 | .destroy = prio_destroy, |
| 460 | .change = prio_tune, |
| 461 | .dump = prio_dump, |
| 462 | .owner = THIS_MODULE, |
| 463 | }; |
| 464 | |
| 465 | static int __init prio_module_init(void) |
| 466 | { |
| 467 | return register_qdisc(&prio_qdisc_ops); |
| 468 | } |
| 469 | |
| 470 | static void __exit prio_module_exit(void) |
| 471 | { |
| 472 | unregister_qdisc(&prio_qdisc_ops); |
| 473 | } |
| 474 | |
| 475 | module_init(prio_module_init) |
| 476 | module_exit(prio_module_exit) |
| 477 | |
| 478 | MODULE_LICENSE("GPL"); |