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, see <http://www.gnu.org/licenses/>. 15 * 16 * Author: Alexander Duyck <alexander.h.duyck@intel.com> 17 */ 18 19 #include <linux/module.h> 20 #include <linux/slab.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 __rcu *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 struct tcf_proto *fl = rcu_dereference_bh(q->filter_list); 46 int err; 47 48 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 49 err = tc_classify(skb, fl, &res); 50 #ifdef CONFIG_NET_CLS_ACT 51 switch (err) { 52 case TC_ACT_STOLEN: 53 case TC_ACT_QUEUED: 54 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 55 case TC_ACT_SHOT: 56 return NULL; 57 } 58 #endif 59 band = skb_get_queue_mapping(skb); 60 61 if (band >= q->bands) 62 return q->queues[0]; 63 64 return q->queues[band]; 65 } 66 67 static int 68 multiq_enqueue(struct sk_buff *skb, struct Qdisc *sch) 69 { 70 struct Qdisc *qdisc; 71 int ret; 72 73 qdisc = multiq_classify(skb, sch, &ret); 74 #ifdef CONFIG_NET_CLS_ACT 75 if (qdisc == NULL) { 76 77 if (ret & __NET_XMIT_BYPASS) 78 qdisc_qstats_drop(sch); 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 qdisc_qstats_drop(sch); 91 return ret; 92 } 93 94 static struct sk_buff *multiq_dequeue(struct Qdisc *sch) 95 { 96 struct multiq_sched_data *q = qdisc_priv(sch); 97 struct Qdisc *qdisc; 98 struct sk_buff *skb; 99 int band; 100 101 for (band = 0; band < q->bands; band++) { 102 /* cycle through bands to ensure fairness */ 103 q->curband++; 104 if (q->curband >= q->bands) 105 q->curband = 0; 106 107 /* Check that target subqueue is available before 108 * pulling an skb to avoid head-of-line blocking. 109 */ 110 if (!netif_xmit_stopped( 111 netdev_get_tx_queue(qdisc_dev(sch), q->curband))) { 112 qdisc = q->queues[q->curband]; 113 skb = qdisc->dequeue(qdisc); 114 if (skb) { 115 qdisc_bstats_update(sch, skb); 116 sch->q.qlen--; 117 return skb; 118 } 119 } 120 } 121 return NULL; 122 123 } 124 125 static struct sk_buff *multiq_peek(struct Qdisc *sch) 126 { 127 struct multiq_sched_data *q = qdisc_priv(sch); 128 unsigned int curband = q->curband; 129 struct Qdisc *qdisc; 130 struct sk_buff *skb; 131 int band; 132 133 for (band = 0; band < q->bands; band++) { 134 /* cycle through bands to ensure fairness */ 135 curband++; 136 if (curband >= q->bands) 137 curband = 0; 138 139 /* Check that target subqueue is available before 140 * pulling an skb to avoid head-of-line blocking. 141 */ 142 if (!netif_xmit_stopped( 143 netdev_get_tx_queue(qdisc_dev(sch), curband))) { 144 qdisc = q->queues[curband]; 145 skb = qdisc->ops->peek(qdisc); 146 if (skb) 147 return skb; 148 } 149 } 150 return NULL; 151 152 } 153 154 static unsigned int multiq_drop(struct Qdisc *sch) 155 { 156 struct multiq_sched_data *q = qdisc_priv(sch); 157 int band; 158 unsigned int len; 159 struct Qdisc *qdisc; 160 161 for (band = q->bands - 1; band >= 0; band--) { 162 qdisc = q->queues[band]; 163 if (qdisc->ops->drop) { 164 len = qdisc->ops->drop(qdisc); 165 if (len != 0) { 166 sch->q.qlen--; 167 return len; 168 } 169 } 170 } 171 return 0; 172 } 173 174 175 static void 176 multiq_reset(struct Qdisc *sch) 177 { 178 u16 band; 179 struct multiq_sched_data *q = qdisc_priv(sch); 180 181 for (band = 0; band < q->bands; band++) 182 qdisc_reset(q->queues[band]); 183 sch->q.qlen = 0; 184 q->curband = 0; 185 } 186 187 static void 188 multiq_destroy(struct Qdisc *sch) 189 { 190 int band; 191 struct multiq_sched_data *q = qdisc_priv(sch); 192 193 tcf_destroy_chain(&q->filter_list); 194 for (band = 0; band < q->bands; band++) 195 qdisc_destroy(q->queues[band]); 196 197 kfree(q->queues); 198 } 199 200 static int multiq_tune(struct Qdisc *sch, struct nlattr *opt) 201 { 202 struct multiq_sched_data *q = qdisc_priv(sch); 203 struct tc_multiq_qopt *qopt; 204 int i; 205 206 if (!netif_is_multiqueue(qdisc_dev(sch))) 207 return -EOPNOTSUPP; 208 if (nla_len(opt) < sizeof(*qopt)) 209 return -EINVAL; 210 211 qopt = nla_data(opt); 212 213 qopt->bands = qdisc_dev(sch)->real_num_tx_queues; 214 215 sch_tree_lock(sch); 216 q->bands = qopt->bands; 217 for (i = q->bands; i < q->max_bands; i++) { 218 if (q->queues[i] != &noop_qdisc) { 219 struct Qdisc *child = q->queues[i]; 220 q->queues[i] = &noop_qdisc; 221 qdisc_tree_decrease_qlen(child, child->q.qlen); 222 qdisc_destroy(child); 223 } 224 } 225 226 sch_tree_unlock(sch); 227 228 for (i = 0; i < q->bands; i++) { 229 if (q->queues[i] == &noop_qdisc) { 230 struct Qdisc *child, *old; 231 child = qdisc_create_dflt(sch->dev_queue, 232 &pfifo_qdisc_ops, 233 TC_H_MAKE(sch->handle, 234 i + 1)); 235 if (child) { 236 sch_tree_lock(sch); 237 old = q->queues[i]; 238 q->queues[i] = child; 239 240 if (old != &noop_qdisc) { 241 qdisc_tree_decrease_qlen(old, 242 old->q.qlen); 243 qdisc_destroy(old); 244 } 245 sch_tree_unlock(sch); 246 } 247 } 248 } 249 return 0; 250 } 251 252 static int multiq_init(struct Qdisc *sch, struct nlattr *opt) 253 { 254 struct multiq_sched_data *q = qdisc_priv(sch); 255 int i, err; 256 257 q->queues = NULL; 258 259 if (opt == NULL) 260 return -EINVAL; 261 262 q->max_bands = qdisc_dev(sch)->num_tx_queues; 263 264 q->queues = kcalloc(q->max_bands, sizeof(struct Qdisc *), GFP_KERNEL); 265 if (!q->queues) 266 return -ENOBUFS; 267 for (i = 0; i < q->max_bands; i++) 268 q->queues[i] = &noop_qdisc; 269 270 err = multiq_tune(sch, opt); 271 272 if (err) 273 kfree(q->queues); 274 275 return err; 276 } 277 278 static int multiq_dump(struct Qdisc *sch, struct sk_buff *skb) 279 { 280 struct multiq_sched_data *q = qdisc_priv(sch); 281 unsigned char *b = skb_tail_pointer(skb); 282 struct tc_multiq_qopt opt; 283 284 opt.bands = q->bands; 285 opt.max_bands = q->max_bands; 286 287 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) 288 goto nla_put_failure; 289 290 return skb->len; 291 292 nla_put_failure: 293 nlmsg_trim(skb, b); 294 return -1; 295 } 296 297 static int multiq_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 298 struct Qdisc **old) 299 { 300 struct multiq_sched_data *q = qdisc_priv(sch); 301 unsigned long band = arg - 1; 302 303 if (new == NULL) 304 new = &noop_qdisc; 305 306 sch_tree_lock(sch); 307 *old = q->queues[band]; 308 q->queues[band] = new; 309 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); 310 qdisc_reset(*old); 311 sch_tree_unlock(sch); 312 313 return 0; 314 } 315 316 static struct Qdisc * 317 multiq_leaf(struct Qdisc *sch, unsigned long arg) 318 { 319 struct multiq_sched_data *q = qdisc_priv(sch); 320 unsigned long band = arg - 1; 321 322 return q->queues[band]; 323 } 324 325 static unsigned long multiq_get(struct Qdisc *sch, u32 classid) 326 { 327 struct multiq_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 multiq_bind(struct Qdisc *sch, unsigned long parent, 336 u32 classid) 337 { 338 return multiq_get(sch, classid); 339 } 340 341 342 static void multiq_put(struct Qdisc *q, unsigned long cl) 343 { 344 } 345 346 static int multiq_dump_class(struct Qdisc *sch, unsigned long cl, 347 struct sk_buff *skb, struct tcmsg *tcm) 348 { 349 struct multiq_sched_data *q = qdisc_priv(sch); 350 351 tcm->tcm_handle |= TC_H_MIN(cl); 352 tcm->tcm_info = q->queues[cl - 1]->handle; 353 return 0; 354 } 355 356 static int multiq_dump_class_stats(struct Qdisc *sch, unsigned long cl, 357 struct gnet_dump *d) 358 { 359 struct multiq_sched_data *q = qdisc_priv(sch); 360 struct Qdisc *cl_q; 361 362 cl_q = q->queues[cl - 1]; 363 if (gnet_stats_copy_basic(d, NULL, &cl_q->bstats) < 0 || 364 gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0) 365 return -1; 366 367 return 0; 368 } 369 370 static void multiq_walk(struct Qdisc *sch, struct qdisc_walker *arg) 371 { 372 struct multiq_sched_data *q = qdisc_priv(sch); 373 int band; 374 375 if (arg->stop) 376 return; 377 378 for (band = 0; band < q->bands; band++) { 379 if (arg->count < arg->skip) { 380 arg->count++; 381 continue; 382 } 383 if (arg->fn(sch, band + 1, arg) < 0) { 384 arg->stop = 1; 385 break; 386 } 387 arg->count++; 388 } 389 } 390 391 static struct tcf_proto __rcu **multiq_find_tcf(struct Qdisc *sch, 392 unsigned long cl) 393 { 394 struct multiq_sched_data *q = qdisc_priv(sch); 395 396 if (cl) 397 return NULL; 398 return &q->filter_list; 399 } 400 401 static const struct Qdisc_class_ops multiq_class_ops = { 402 .graft = multiq_graft, 403 .leaf = multiq_leaf, 404 .get = multiq_get, 405 .put = multiq_put, 406 .walk = multiq_walk, 407 .tcf_chain = multiq_find_tcf, 408 .bind_tcf = multiq_bind, 409 .unbind_tcf = multiq_put, 410 .dump = multiq_dump_class, 411 .dump_stats = multiq_dump_class_stats, 412 }; 413 414 static struct Qdisc_ops multiq_qdisc_ops __read_mostly = { 415 .next = NULL, 416 .cl_ops = &multiq_class_ops, 417 .id = "multiq", 418 .priv_size = sizeof(struct multiq_sched_data), 419 .enqueue = multiq_enqueue, 420 .dequeue = multiq_dequeue, 421 .peek = multiq_peek, 422 .drop = multiq_drop, 423 .init = multiq_init, 424 .reset = multiq_reset, 425 .destroy = multiq_destroy, 426 .change = multiq_tune, 427 .dump = multiq_dump, 428 .owner = THIS_MODULE, 429 }; 430 431 static int __init multiq_module_init(void) 432 { 433 return register_qdisc(&multiq_qdisc_ops); 434 } 435 436 static void __exit multiq_module_exit(void) 437 { 438 unregister_qdisc(&multiq_qdisc_ops); 439 } 440 441 module_init(multiq_module_init) 442 module_exit(multiq_module_exit) 443 444 MODULE_LICENSE("GPL"); 445