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, false); 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 void 155 multiq_reset(struct Qdisc *sch) 156 { 157 u16 band; 158 struct multiq_sched_data *q = qdisc_priv(sch); 159 160 for (band = 0; band < q->bands; band++) 161 qdisc_reset(q->queues[band]); 162 sch->q.qlen = 0; 163 q->curband = 0; 164 } 165 166 static void 167 multiq_destroy(struct Qdisc *sch) 168 { 169 int band; 170 struct multiq_sched_data *q = qdisc_priv(sch); 171 172 tcf_destroy_chain(&q->filter_list); 173 for (band = 0; band < q->bands; band++) 174 qdisc_destroy(q->queues[band]); 175 176 kfree(q->queues); 177 } 178 179 static int multiq_tune(struct Qdisc *sch, struct nlattr *opt) 180 { 181 struct multiq_sched_data *q = qdisc_priv(sch); 182 struct tc_multiq_qopt *qopt; 183 int i; 184 185 if (!netif_is_multiqueue(qdisc_dev(sch))) 186 return -EOPNOTSUPP; 187 if (nla_len(opt) < sizeof(*qopt)) 188 return -EINVAL; 189 190 qopt = nla_data(opt); 191 192 qopt->bands = qdisc_dev(sch)->real_num_tx_queues; 193 194 sch_tree_lock(sch); 195 q->bands = qopt->bands; 196 for (i = q->bands; i < q->max_bands; i++) { 197 if (q->queues[i] != &noop_qdisc) { 198 struct Qdisc *child = q->queues[i]; 199 q->queues[i] = &noop_qdisc; 200 qdisc_tree_reduce_backlog(child, child->q.qlen, 201 child->qstats.backlog); 202 qdisc_destroy(child); 203 } 204 } 205 206 sch_tree_unlock(sch); 207 208 for (i = 0; i < q->bands; i++) { 209 if (q->queues[i] == &noop_qdisc) { 210 struct Qdisc *child, *old; 211 child = qdisc_create_dflt(sch->dev_queue, 212 &pfifo_qdisc_ops, 213 TC_H_MAKE(sch->handle, 214 i + 1)); 215 if (child) { 216 sch_tree_lock(sch); 217 old = q->queues[i]; 218 q->queues[i] = child; 219 220 if (old != &noop_qdisc) { 221 qdisc_tree_reduce_backlog(old, 222 old->q.qlen, 223 old->qstats.backlog); 224 qdisc_destroy(old); 225 } 226 sch_tree_unlock(sch); 227 } 228 } 229 } 230 return 0; 231 } 232 233 static int multiq_init(struct Qdisc *sch, struct nlattr *opt) 234 { 235 struct multiq_sched_data *q = qdisc_priv(sch); 236 int i, err; 237 238 q->queues = NULL; 239 240 if (opt == NULL) 241 return -EINVAL; 242 243 q->max_bands = qdisc_dev(sch)->num_tx_queues; 244 245 q->queues = kcalloc(q->max_bands, sizeof(struct Qdisc *), GFP_KERNEL); 246 if (!q->queues) 247 return -ENOBUFS; 248 for (i = 0; i < q->max_bands; i++) 249 q->queues[i] = &noop_qdisc; 250 251 err = multiq_tune(sch, opt); 252 253 if (err) 254 kfree(q->queues); 255 256 return err; 257 } 258 259 static int multiq_dump(struct Qdisc *sch, struct sk_buff *skb) 260 { 261 struct multiq_sched_data *q = qdisc_priv(sch); 262 unsigned char *b = skb_tail_pointer(skb); 263 struct tc_multiq_qopt opt; 264 265 opt.bands = q->bands; 266 opt.max_bands = q->max_bands; 267 268 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) 269 goto nla_put_failure; 270 271 return skb->len; 272 273 nla_put_failure: 274 nlmsg_trim(skb, b); 275 return -1; 276 } 277 278 static int multiq_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 279 struct Qdisc **old) 280 { 281 struct multiq_sched_data *q = qdisc_priv(sch); 282 unsigned long band = arg - 1; 283 284 if (new == NULL) 285 new = &noop_qdisc; 286 287 *old = qdisc_replace(sch, new, &q->queues[band]); 288 return 0; 289 } 290 291 static struct Qdisc * 292 multiq_leaf(struct Qdisc *sch, unsigned long arg) 293 { 294 struct multiq_sched_data *q = qdisc_priv(sch); 295 unsigned long band = arg - 1; 296 297 return q->queues[band]; 298 } 299 300 static unsigned long multiq_get(struct Qdisc *sch, u32 classid) 301 { 302 struct multiq_sched_data *q = qdisc_priv(sch); 303 unsigned long band = TC_H_MIN(classid); 304 305 if (band - 1 >= q->bands) 306 return 0; 307 return band; 308 } 309 310 static unsigned long multiq_bind(struct Qdisc *sch, unsigned long parent, 311 u32 classid) 312 { 313 return multiq_get(sch, classid); 314 } 315 316 317 static void multiq_put(struct Qdisc *q, unsigned long cl) 318 { 319 } 320 321 static int multiq_dump_class(struct Qdisc *sch, unsigned long cl, 322 struct sk_buff *skb, struct tcmsg *tcm) 323 { 324 struct multiq_sched_data *q = qdisc_priv(sch); 325 326 tcm->tcm_handle |= TC_H_MIN(cl); 327 tcm->tcm_info = q->queues[cl - 1]->handle; 328 return 0; 329 } 330 331 static int multiq_dump_class_stats(struct Qdisc *sch, unsigned long cl, 332 struct gnet_dump *d) 333 { 334 struct multiq_sched_data *q = qdisc_priv(sch); 335 struct Qdisc *cl_q; 336 337 cl_q = q->queues[cl - 1]; 338 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), 339 d, NULL, &cl_q->bstats) < 0 || 340 gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0) 341 return -1; 342 343 return 0; 344 } 345 346 static void multiq_walk(struct Qdisc *sch, struct qdisc_walker *arg) 347 { 348 struct multiq_sched_data *q = qdisc_priv(sch); 349 int band; 350 351 if (arg->stop) 352 return; 353 354 for (band = 0; band < q->bands; band++) { 355 if (arg->count < arg->skip) { 356 arg->count++; 357 continue; 358 } 359 if (arg->fn(sch, band + 1, arg) < 0) { 360 arg->stop = 1; 361 break; 362 } 363 arg->count++; 364 } 365 } 366 367 static struct tcf_proto __rcu **multiq_find_tcf(struct Qdisc *sch, 368 unsigned long cl) 369 { 370 struct multiq_sched_data *q = qdisc_priv(sch); 371 372 if (cl) 373 return NULL; 374 return &q->filter_list; 375 } 376 377 static const struct Qdisc_class_ops multiq_class_ops = { 378 .graft = multiq_graft, 379 .leaf = multiq_leaf, 380 .get = multiq_get, 381 .put = multiq_put, 382 .walk = multiq_walk, 383 .tcf_chain = multiq_find_tcf, 384 .bind_tcf = multiq_bind, 385 .unbind_tcf = multiq_put, 386 .dump = multiq_dump_class, 387 .dump_stats = multiq_dump_class_stats, 388 }; 389 390 static struct Qdisc_ops multiq_qdisc_ops __read_mostly = { 391 .next = NULL, 392 .cl_ops = &multiq_class_ops, 393 .id = "multiq", 394 .priv_size = sizeof(struct multiq_sched_data), 395 .enqueue = multiq_enqueue, 396 .dequeue = multiq_dequeue, 397 .peek = multiq_peek, 398 .init = multiq_init, 399 .reset = multiq_reset, 400 .destroy = multiq_destroy, 401 .change = multiq_tune, 402 .dump = multiq_dump, 403 .owner = THIS_MODULE, 404 }; 405 406 static int __init multiq_module_init(void) 407 { 408 return register_qdisc(&multiq_qdisc_ops); 409 } 410 411 static void __exit multiq_module_exit(void) 412 { 413 unregister_qdisc(&multiq_qdisc_ops); 414 } 415 416 module_init(multiq_module_init) 417 module_exit(multiq_module_exit) 418 419 MODULE_LICENSE("GPL"); 420