1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * net/sched/sch_mq.c Classful multiqueue dummy scheduler 4 * 5 * Copyright (c) 2009 Patrick McHardy <kaber@trash.net> 6 */ 7 8 #include <linux/types.h> 9 #include <linux/slab.h> 10 #include <linux/kernel.h> 11 #include <linux/export.h> 12 #include <linux/string.h> 13 #include <linux/errno.h> 14 #include <linux/skbuff.h> 15 #include <net/netlink.h> 16 #include <net/pkt_cls.h> 17 #include <net/pkt_sched.h> 18 #include <net/sch_priv.h> 19 20 static int mq_offload(struct Qdisc *sch, enum tc_mq_command cmd) 21 { 22 struct net_device *dev = qdisc_dev(sch); 23 struct tc_mq_qopt_offload opt = { 24 .command = cmd, 25 .handle = sch->handle, 26 }; 27 28 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 29 return -EOPNOTSUPP; 30 31 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_MQ, &opt); 32 } 33 34 static int mq_offload_stats(struct Qdisc *sch) 35 { 36 struct tc_mq_qopt_offload opt = { 37 .command = TC_MQ_STATS, 38 .handle = sch->handle, 39 .stats = { 40 .bstats = &sch->bstats, 41 .qstats = &sch->qstats, 42 }, 43 }; 44 45 return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_MQ, &opt); 46 } 47 48 void mq_destroy_common(struct Qdisc *sch) 49 { 50 struct net_device *dev = qdisc_dev(sch); 51 struct mq_sched *priv = qdisc_priv(sch); 52 unsigned int ntx; 53 54 if (!priv->qdiscs) 55 return; 56 for (ntx = 0; ntx < dev->num_tx_queues && priv->qdiscs[ntx]; ntx++) 57 qdisc_put(priv->qdiscs[ntx]); 58 kfree(priv->qdiscs); 59 } 60 EXPORT_SYMBOL_NS_GPL(mq_destroy_common, "NET_SCHED_INTERNAL"); 61 62 static void mq_destroy(struct Qdisc *sch) 63 { 64 mq_offload(sch, TC_MQ_DESTROY); 65 mq_destroy_common(sch); 66 } 67 68 int mq_init_common(struct Qdisc *sch, struct nlattr *opt, 69 struct netlink_ext_ack *extack, 70 const struct Qdisc_ops *qdisc_ops) 71 { 72 struct net_device *dev = qdisc_dev(sch); 73 struct mq_sched *priv = qdisc_priv(sch); 74 struct netdev_queue *dev_queue; 75 struct Qdisc *qdisc; 76 unsigned int ntx; 77 78 if (sch->parent != TC_H_ROOT) 79 return -EOPNOTSUPP; 80 81 if (!netif_is_multiqueue(dev)) 82 return -EOPNOTSUPP; 83 84 /* pre-allocate qdiscs, attachment can't fail */ 85 priv->qdiscs = kzalloc_objs(priv->qdiscs[0], dev->num_tx_queues); 86 if (!priv->qdiscs) 87 return -ENOMEM; 88 89 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) { 90 dev_queue = netdev_get_tx_queue(dev, ntx); 91 qdisc = qdisc_create_dflt(dev_queue, 92 qdisc_ops ?: get_default_qdisc_ops(dev, ntx), 93 TC_H_MAKE(TC_H_MAJ(sch->handle), 94 TC_H_MIN(ntx + 1)), 95 extack); 96 if (!qdisc) 97 return -ENOMEM; 98 priv->qdiscs[ntx] = qdisc; 99 qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT; 100 } 101 102 sch->flags |= TCQ_F_MQROOT; 103 return 0; 104 } 105 EXPORT_SYMBOL_NS_GPL(mq_init_common, "NET_SCHED_INTERNAL"); 106 107 static int mq_init(struct Qdisc *sch, struct nlattr *opt, 108 struct netlink_ext_ack *extack) 109 { 110 int ret; 111 112 ret = mq_init_common(sch, opt, extack, NULL); 113 if (ret) 114 return ret; 115 116 mq_offload(sch, TC_MQ_CREATE); 117 return 0; 118 } 119 120 void mq_attach(struct Qdisc *sch) 121 { 122 struct net_device *dev = qdisc_dev(sch); 123 struct mq_sched *priv = qdisc_priv(sch); 124 struct Qdisc *qdisc, *old; 125 unsigned int ntx; 126 127 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) { 128 qdisc = priv->qdiscs[ntx]; 129 old = dev_graft_qdisc(qdisc->dev_queue, qdisc); 130 if (old) 131 qdisc_put(old); 132 #ifdef CONFIG_NET_SCHED 133 if (ntx < dev->real_num_tx_queues) 134 qdisc_hash_add(qdisc, false); 135 #endif 136 137 } 138 kfree(priv->qdiscs); 139 priv->qdiscs = NULL; 140 } 141 EXPORT_SYMBOL_NS_GPL(mq_attach, "NET_SCHED_INTERNAL"); 142 143 void mq_dump_common(struct Qdisc *sch, struct sk_buff *skb) 144 { 145 struct net_device *dev = qdisc_dev(sch); 146 struct gnet_stats_queue qstats = { 0 }; 147 struct gnet_stats_basic_sync bstats; 148 const struct Qdisc *qdisc; 149 unsigned int qlen = 0; 150 unsigned int ntx; 151 152 gnet_stats_basic_sync_init(&bstats); 153 154 /* MQ supports lockless qdiscs. However, statistics accounting needs 155 * to account for all, none, or a mix of locked and unlocked child 156 * qdiscs. Percpu stats are added to counters in-band and locking 157 * qdisc totals are added at end. 158 */ 159 rcu_read_lock(); 160 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) { 161 qdisc = rcu_dereference(netdev_get_tx_queue(dev, ntx)->qdisc_sleeping); 162 163 gnet_stats_add_basic(&bstats, qdisc->cpu_bstats, 164 &qdisc->bstats, true); 165 gnet_stats_add_queue(&qstats, qdisc->cpu_qstats, 166 &qdisc->qstats); 167 qlen += qdisc_qlen_lockless(qdisc); 168 } 169 rcu_read_unlock(); 170 171 spin_lock_bh(qdisc_lock(sch)); 172 _bstats_set(&sch->bstats, u64_stats_read(&bstats.bytes), 173 u64_stats_read(&bstats.packets)); 174 spin_unlock_bh(qdisc_lock(sch)); 175 176 WRITE_ONCE(sch->qstats.qlen, qstats.qlen); 177 WRITE_ONCE(sch->qstats.backlog, qstats.backlog); 178 WRITE_ONCE(sch->qstats.drops, qstats.drops); 179 WRITE_ONCE(sch->qstats.requeues, qstats.requeues); 180 WRITE_ONCE(sch->qstats.overlimits, qstats.overlimits); 181 182 WRITE_ONCE(sch->q.qlen, qlen); 183 } 184 EXPORT_SYMBOL_NS_GPL(mq_dump_common, "NET_SCHED_INTERNAL"); 185 186 static int mq_dump(struct Qdisc *sch, struct sk_buff *skb) 187 { 188 mq_dump_common(sch, skb); 189 return mq_offload_stats(sch); 190 } 191 192 static struct netdev_queue *mq_queue_get(struct Qdisc *sch, unsigned long cl) 193 { 194 struct net_device *dev = qdisc_dev(sch); 195 unsigned long ntx = cl - 1; 196 197 if (ntx >= dev->num_tx_queues) 198 return NULL; 199 return netdev_get_tx_queue(dev, ntx); 200 } 201 202 struct netdev_queue *mq_select_queue(struct Qdisc *sch, 203 struct tcmsg *tcm) 204 { 205 return mq_queue_get(sch, TC_H_MIN(tcm->tcm_parent)); 206 } 207 EXPORT_SYMBOL_NS_GPL(mq_select_queue, "NET_SCHED_INTERNAL"); 208 209 static int mq_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new, 210 struct Qdisc **old, struct netlink_ext_ack *extack) 211 { 212 struct netdev_queue *dev_queue = mq_queue_get(sch, cl); 213 struct tc_mq_qopt_offload graft_offload; 214 struct net_device *dev = qdisc_dev(sch); 215 216 if (dev->flags & IFF_UP) 217 dev_deactivate(dev, false); 218 219 *old = dev_graft_qdisc(dev_queue, new); 220 if (new) 221 new->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT; 222 if (dev->flags & IFF_UP) 223 dev_activate(dev); 224 225 graft_offload.handle = sch->handle; 226 graft_offload.graft_params.queue = cl - 1; 227 graft_offload.graft_params.child_handle = new ? new->handle : 0; 228 graft_offload.command = TC_MQ_GRAFT; 229 230 qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, *old, 231 TC_SETUP_QDISC_MQ, &graft_offload, extack); 232 return 0; 233 } 234 235 struct Qdisc *mq_leaf(struct Qdisc *sch, unsigned long cl) 236 { 237 struct netdev_queue *dev_queue = mq_queue_get(sch, cl); 238 239 return rtnl_dereference(dev_queue->qdisc_sleeping); 240 } 241 EXPORT_SYMBOL_NS_GPL(mq_leaf, "NET_SCHED_INTERNAL"); 242 243 unsigned long mq_find(struct Qdisc *sch, u32 classid) 244 { 245 unsigned int ntx = TC_H_MIN(classid); 246 247 if (!mq_queue_get(sch, ntx)) 248 return 0; 249 return ntx; 250 } 251 EXPORT_SYMBOL_NS_GPL(mq_find, "NET_SCHED_INTERNAL"); 252 253 int mq_dump_class(struct Qdisc *sch, unsigned long cl, 254 struct sk_buff *skb, struct tcmsg *tcm) 255 { 256 struct netdev_queue *dev_queue = mq_queue_get(sch, cl); 257 258 tcm->tcm_parent = TC_H_ROOT; 259 tcm->tcm_handle |= TC_H_MIN(cl); 260 tcm->tcm_info = rtnl_dereference(dev_queue->qdisc_sleeping)->handle; 261 return 0; 262 } 263 EXPORT_SYMBOL_NS_GPL(mq_dump_class, "NET_SCHED_INTERNAL"); 264 265 int mq_dump_class_stats(struct Qdisc *sch, unsigned long cl, 266 struct gnet_dump *d) 267 { 268 struct netdev_queue *dev_queue = mq_queue_get(sch, cl); 269 270 sch = rtnl_dereference(dev_queue->qdisc_sleeping); 271 if (gnet_stats_copy_basic(d, sch->cpu_bstats, &sch->bstats, true) < 0 || 272 qdisc_qstats_copy(d, sch) < 0) 273 return -1; 274 return 0; 275 } 276 EXPORT_SYMBOL_NS_GPL(mq_dump_class_stats, "NET_SCHED_INTERNAL"); 277 278 void mq_walk(struct Qdisc *sch, struct qdisc_walker *arg) 279 { 280 struct net_device *dev = qdisc_dev(sch); 281 unsigned int ntx; 282 283 if (arg->stop) 284 return; 285 286 arg->count = arg->skip; 287 for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) { 288 if (!tc_qdisc_stats_dump(sch, ntx + 1, arg)) 289 break; 290 } 291 } 292 EXPORT_SYMBOL_NS_GPL(mq_walk, "NET_SCHED_INTERNAL"); 293 294 static const struct Qdisc_class_ops mq_class_ops = { 295 .select_queue = mq_select_queue, 296 .graft = mq_graft, 297 .leaf = mq_leaf, 298 .find = mq_find, 299 .walk = mq_walk, 300 .dump = mq_dump_class, 301 .dump_stats = mq_dump_class_stats, 302 }; 303 304 struct Qdisc_ops mq_qdisc_ops __read_mostly = { 305 .cl_ops = &mq_class_ops, 306 .id = "mq", 307 .priv_size = sizeof(struct mq_sched), 308 .init = mq_init, 309 .destroy = mq_destroy, 310 .attach = mq_attach, 311 .change_real_num_tx = mq_change_real_num_tx, 312 .dump = mq_dump, 313 .owner = THIS_MODULE, 314 }; 315