1 /* 2 * Copyright (c) 2008, Intel Corporation. 3 * 4 * This program is free software; you can redistribute it and/or modify it 5 * under the terms and conditions of the GNU General Public License, 6 * version 2, as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope it will be useful, but WITHOUT 9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 11 * more details. 12 * 13 * You should have received a copy of the GNU General Public License along with 14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple 15 * Place - Suite 330, Boston, MA 02111-1307 USA. 16 * 17 * Author: Alexander Duyck <alexander.h.duyck@intel.com> 18 */ 19 20 #include <linux/module.h> 21 #include <linux/types.h> 22 #include <linux/kernel.h> 23 #include <linux/string.h> 24 #include <linux/errno.h> 25 #include <linux/skbuff.h> 26 #include <net/netlink.h> 27 #include <net/pkt_sched.h> 28 29 30 struct multiq_sched_data { 31 u16 bands; 32 u16 max_bands; 33 u16 curband; 34 struct tcf_proto *filter_list; 35 struct Qdisc **queues; 36 }; 37 38 39 static struct Qdisc * 40 multiq_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr) 41 { 42 struct multiq_sched_data *q = qdisc_priv(sch); 43 u32 band; 44 struct tcf_result res; 45 int err; 46 47 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 48 err = tc_classify(skb, q->filter_list, &res); 49 #ifdef CONFIG_NET_CLS_ACT 50 switch (err) { 51 case TC_ACT_STOLEN: 52 case TC_ACT_QUEUED: 53 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 54 case TC_ACT_SHOT: 55 return NULL; 56 } 57 #endif 58 band = skb_get_queue_mapping(skb); 59 60 if (band >= q->bands) 61 return q->queues[0]; 62 63 return q->queues[band]; 64 } 65 66 static int 67 multiq_enqueue(struct sk_buff *skb, struct Qdisc *sch) 68 { 69 struct Qdisc *qdisc; 70 int ret; 71 72 qdisc = multiq_classify(skb, sch, &ret); 73 #ifdef CONFIG_NET_CLS_ACT 74 if (qdisc == NULL) { 75 76 if (ret & __NET_XMIT_BYPASS) 77 sch->qstats.drops++; 78 kfree_skb(skb); 79 return ret; 80 } 81 #endif 82 83 ret = qdisc_enqueue(skb, qdisc); 84 if (ret == NET_XMIT_SUCCESS) { 85 sch->bstats.bytes += qdisc_pkt_len(skb); 86 sch->bstats.packets++; 87 sch->q.qlen++; 88 return NET_XMIT_SUCCESS; 89 } 90 if (net_xmit_drop_count(ret)) 91 sch->qstats.drops++; 92 return ret; 93 } 94 95 96 static int 97 multiq_requeue(struct sk_buff *skb, struct Qdisc *sch) 98 { 99 struct Qdisc *qdisc; 100 struct multiq_sched_data *q = qdisc_priv(sch); 101 int ret; 102 103 qdisc = multiq_classify(skb, sch, &ret); 104 #ifdef CONFIG_NET_CLS_ACT 105 if (qdisc == NULL) { 106 if (ret & __NET_XMIT_BYPASS) 107 sch->qstats.drops++; 108 kfree_skb(skb); 109 return ret; 110 } 111 #endif 112 113 ret = qdisc->ops->requeue(skb, qdisc); 114 if (ret == NET_XMIT_SUCCESS) { 115 sch->q.qlen++; 116 sch->qstats.requeues++; 117 if (q->curband) 118 q->curband--; 119 else 120 q->curband = q->bands - 1; 121 return NET_XMIT_SUCCESS; 122 } 123 if (net_xmit_drop_count(ret)) 124 sch->qstats.drops++; 125 return ret; 126 } 127 128 129 static struct sk_buff *multiq_dequeue(struct Qdisc *sch) 130 { 131 struct multiq_sched_data *q = qdisc_priv(sch); 132 struct Qdisc *qdisc; 133 struct sk_buff *skb; 134 int band; 135 136 for (band = 0; band < q->bands; band++) { 137 /* cycle through bands to ensure fairness */ 138 q->curband++; 139 if (q->curband >= q->bands) 140 q->curband = 0; 141 142 /* Check that target subqueue is available before 143 * pulling an skb to avoid excessive requeues 144 */ 145 if (!__netif_subqueue_stopped(qdisc_dev(sch), q->curband)) { 146 qdisc = q->queues[q->curband]; 147 skb = qdisc->dequeue(qdisc); 148 if (skb) { 149 sch->q.qlen--; 150 return skb; 151 } 152 } 153 } 154 return NULL; 155 156 } 157 158 static unsigned int multiq_drop(struct Qdisc *sch) 159 { 160 struct multiq_sched_data *q = qdisc_priv(sch); 161 int band; 162 unsigned int len; 163 struct Qdisc *qdisc; 164 165 for (band = q->bands-1; band >= 0; band--) { 166 qdisc = q->queues[band]; 167 if (qdisc->ops->drop) { 168 len = qdisc->ops->drop(qdisc); 169 if (len != 0) { 170 sch->q.qlen--; 171 return len; 172 } 173 } 174 } 175 return 0; 176 } 177 178 179 static void 180 multiq_reset(struct Qdisc *sch) 181 { 182 u16 band; 183 struct multiq_sched_data *q = qdisc_priv(sch); 184 185 for (band = 0; band < q->bands; band++) 186 qdisc_reset(q->queues[band]); 187 sch->q.qlen = 0; 188 q->curband = 0; 189 } 190 191 static void 192 multiq_destroy(struct Qdisc *sch) 193 { 194 int band; 195 struct multiq_sched_data *q = qdisc_priv(sch); 196 197 tcf_destroy_chain(&q->filter_list); 198 for (band = 0; band < q->bands; band++) 199 qdisc_destroy(q->queues[band]); 200 201 kfree(q->queues); 202 } 203 204 static int multiq_tune(struct Qdisc *sch, struct nlattr *opt) 205 { 206 struct multiq_sched_data *q = qdisc_priv(sch); 207 struct tc_multiq_qopt *qopt; 208 int i; 209 210 if (!netif_is_multiqueue(qdisc_dev(sch))) 211 return -EINVAL; 212 if (nla_len(opt) < sizeof(*qopt)) 213 return -EINVAL; 214 215 qopt = nla_data(opt); 216 217 qopt->bands = qdisc_dev(sch)->real_num_tx_queues; 218 219 sch_tree_lock(sch); 220 q->bands = qopt->bands; 221 for (i = q->bands; i < q->max_bands; i++) { 222 if (q->queues[i] != &noop_qdisc) { 223 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc); 224 qdisc_tree_decrease_qlen(child, child->q.qlen); 225 qdisc_destroy(child); 226 } 227 } 228 229 sch_tree_unlock(sch); 230 231 for (i = 0; i < q->bands; i++) { 232 if (q->queues[i] == &noop_qdisc) { 233 struct Qdisc *child; 234 child = qdisc_create_dflt(qdisc_dev(sch), 235 sch->dev_queue, 236 &pfifo_qdisc_ops, 237 TC_H_MAKE(sch->handle, 238 i + 1)); 239 if (child) { 240 sch_tree_lock(sch); 241 child = xchg(&q->queues[i], child); 242 243 if (child != &noop_qdisc) { 244 qdisc_tree_decrease_qlen(child, 245 child->q.qlen); 246 qdisc_destroy(child); 247 } 248 sch_tree_unlock(sch); 249 } 250 } 251 } 252 return 0; 253 } 254 255 static int multiq_init(struct Qdisc *sch, struct nlattr *opt) 256 { 257 struct multiq_sched_data *q = qdisc_priv(sch); 258 int i, err; 259 260 q->queues = NULL; 261 262 if (opt == NULL) 263 return -EINVAL; 264 265 q->max_bands = qdisc_dev(sch)->num_tx_queues; 266 267 q->queues = kcalloc(q->max_bands, sizeof(struct Qdisc *), GFP_KERNEL); 268 if (!q->queues) 269 return -ENOBUFS; 270 for (i = 0; i < q->max_bands; i++) 271 q->queues[i] = &noop_qdisc; 272 273 err = multiq_tune(sch,opt); 274 275 if (err) 276 kfree(q->queues); 277 278 return err; 279 } 280 281 static int multiq_dump(struct Qdisc *sch, struct sk_buff *skb) 282 { 283 struct multiq_sched_data *q = qdisc_priv(sch); 284 unsigned char *b = skb_tail_pointer(skb); 285 struct tc_multiq_qopt opt; 286 287 opt.bands = q->bands; 288 opt.max_bands = q->max_bands; 289 290 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); 291 292 return skb->len; 293 294 nla_put_failure: 295 nlmsg_trim(skb, b); 296 return -1; 297 } 298 299 static int multiq_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 300 struct Qdisc **old) 301 { 302 struct multiq_sched_data *q = qdisc_priv(sch); 303 unsigned long band = arg - 1; 304 305 if (band >= q->bands) 306 return -EINVAL; 307 308 if (new == NULL) 309 new = &noop_qdisc; 310 311 sch_tree_lock(sch); 312 *old = q->queues[band]; 313 q->queues[band] = new; 314 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); 315 qdisc_reset(*old); 316 sch_tree_unlock(sch); 317 318 return 0; 319 } 320 321 static struct Qdisc * 322 multiq_leaf(struct Qdisc *sch, unsigned long arg) 323 { 324 struct multiq_sched_data *q = qdisc_priv(sch); 325 unsigned long band = arg - 1; 326 327 if (band >= q->bands) 328 return NULL; 329 330 return q->queues[band]; 331 } 332 333 static unsigned long multiq_get(struct Qdisc *sch, u32 classid) 334 { 335 struct multiq_sched_data *q = qdisc_priv(sch); 336 unsigned long band = TC_H_MIN(classid); 337 338 if (band - 1 >= q->bands) 339 return 0; 340 return band; 341 } 342 343 static unsigned long multiq_bind(struct Qdisc *sch, unsigned long parent, 344 u32 classid) 345 { 346 return multiq_get(sch, classid); 347 } 348 349 350 static void multiq_put(struct Qdisc *q, unsigned long cl) 351 { 352 return; 353 } 354 355 static int multiq_change(struct Qdisc *sch, u32 handle, u32 parent, 356 struct nlattr **tca, unsigned long *arg) 357 { 358 unsigned long cl = *arg; 359 struct multiq_sched_data *q = qdisc_priv(sch); 360 361 if (cl - 1 > q->bands) 362 return -ENOENT; 363 return 0; 364 } 365 366 static int multiq_delete(struct Qdisc *sch, unsigned long cl) 367 { 368 struct multiq_sched_data *q = qdisc_priv(sch); 369 if (cl - 1 > q->bands) 370 return -ENOENT; 371 return 0; 372 } 373 374 375 static int multiq_dump_class(struct Qdisc *sch, unsigned long cl, 376 struct sk_buff *skb, struct tcmsg *tcm) 377 { 378 struct multiq_sched_data *q = qdisc_priv(sch); 379 380 if (cl - 1 > q->bands) 381 return -ENOENT; 382 tcm->tcm_handle |= TC_H_MIN(cl); 383 if (q->queues[cl-1]) 384 tcm->tcm_info = q->queues[cl-1]->handle; 385 return 0; 386 } 387 388 static int multiq_dump_class_stats(struct Qdisc *sch, unsigned long cl, 389 struct gnet_dump *d) 390 { 391 struct multiq_sched_data *q = qdisc_priv(sch); 392 struct Qdisc *cl_q; 393 394 cl_q = q->queues[cl - 1]; 395 if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 || 396 gnet_stats_copy_queue(d, &cl_q->qstats) < 0) 397 return -1; 398 399 return 0; 400 } 401 402 static void multiq_walk(struct Qdisc *sch, struct qdisc_walker *arg) 403 { 404 struct multiq_sched_data *q = qdisc_priv(sch); 405 int band; 406 407 if (arg->stop) 408 return; 409 410 for (band = 0; band < q->bands; band++) { 411 if (arg->count < arg->skip) { 412 arg->count++; 413 continue; 414 } 415 if (arg->fn(sch, band+1, arg) < 0) { 416 arg->stop = 1; 417 break; 418 } 419 arg->count++; 420 } 421 } 422 423 static struct tcf_proto **multiq_find_tcf(struct Qdisc *sch, unsigned long cl) 424 { 425 struct multiq_sched_data *q = qdisc_priv(sch); 426 427 if (cl) 428 return NULL; 429 return &q->filter_list; 430 } 431 432 static const struct Qdisc_class_ops multiq_class_ops = { 433 .graft = multiq_graft, 434 .leaf = multiq_leaf, 435 .get = multiq_get, 436 .put = multiq_put, 437 .change = multiq_change, 438 .delete = multiq_delete, 439 .walk = multiq_walk, 440 .tcf_chain = multiq_find_tcf, 441 .bind_tcf = multiq_bind, 442 .unbind_tcf = multiq_put, 443 .dump = multiq_dump_class, 444 .dump_stats = multiq_dump_class_stats, 445 }; 446 447 static struct Qdisc_ops multiq_qdisc_ops __read_mostly = { 448 .next = NULL, 449 .cl_ops = &multiq_class_ops, 450 .id = "multiq", 451 .priv_size = sizeof(struct multiq_sched_data), 452 .enqueue = multiq_enqueue, 453 .dequeue = multiq_dequeue, 454 .requeue = multiq_requeue, 455 .drop = multiq_drop, 456 .init = multiq_init, 457 .reset = multiq_reset, 458 .destroy = multiq_destroy, 459 .change = multiq_tune, 460 .dump = multiq_dump, 461 .owner = THIS_MODULE, 462 }; 463 464 static int __init multiq_module_init(void) 465 { 466 return register_qdisc(&multiq_qdisc_ops); 467 } 468 469 static void __exit multiq_module_exit(void) 470 { 471 unregister_qdisc(&multiq_qdisc_ops); 472 } 473 474 module_init(multiq_module_init) 475 module_exit(multiq_module_exit) 476 477 MODULE_LICENSE("GPL"); 478