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