| 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | #ifndef __NET_SCHED_GENERIC_H |
| 3 | #define __NET_SCHED_GENERIC_H |
| 4 | |
| 5 | #include <linux/netdevice.h> |
| 6 | #include <linux/types.h> |
| 7 | #include <linux/rcupdate.h> |
| 8 | #include <linux/pkt_sched.h> |
| 9 | #include <linux/pkt_cls.h> |
| 10 | #include <linux/percpu.h> |
| 11 | #include <linux/dynamic_queue_limits.h> |
| 12 | #include <linux/list.h> |
| 13 | #include <linux/refcount.h> |
| 14 | #include <linux/workqueue.h> |
| 15 | #include <net/gen_stats.h> |
| 16 | #include <net/rtnetlink.h> |
| 17 | |
| 18 | struct Qdisc_ops; |
| 19 | struct qdisc_walker; |
| 20 | struct tcf_walker; |
| 21 | struct module; |
| 22 | |
| 23 | struct qdisc_rate_table { |
| 24 | struct tc_ratespec rate; |
| 25 | u32 data[256]; |
| 26 | struct qdisc_rate_table *next; |
| 27 | int refcnt; |
| 28 | }; |
| 29 | |
| 30 | enum qdisc_state_t { |
| 31 | __QDISC_STATE_SCHED, |
| 32 | __QDISC_STATE_DEACTIVATED, |
| 33 | }; |
| 34 | |
| 35 | struct qdisc_size_table { |
| 36 | struct rcu_head rcu; |
| 37 | struct list_head list; |
| 38 | struct tc_sizespec szopts; |
| 39 | int refcnt; |
| 40 | u16 data[]; |
| 41 | }; |
| 42 | |
| 43 | /* similar to sk_buff_head, but skb->prev pointer is undefined. */ |
| 44 | struct qdisc_skb_head { |
| 45 | struct sk_buff *head; |
| 46 | struct sk_buff *tail; |
| 47 | __u32 qlen; |
| 48 | spinlock_t lock; |
| 49 | }; |
| 50 | |
| 51 | struct Qdisc { |
| 52 | int (*enqueue)(struct sk_buff *skb, |
| 53 | struct Qdisc *sch, |
| 54 | struct sk_buff **to_free); |
| 55 | struct sk_buff * (*dequeue)(struct Qdisc *sch); |
| 56 | unsigned int flags; |
| 57 | #define TCQ_F_BUILTIN 1 |
| 58 | #define TCQ_F_INGRESS 2 |
| 59 | #define TCQ_F_CAN_BYPASS 4 |
| 60 | #define TCQ_F_MQROOT 8 |
| 61 | #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for |
| 62 | * q->dev_queue : It can test |
| 63 | * netif_xmit_frozen_or_stopped() before |
| 64 | * dequeueing next packet. |
| 65 | * Its true for MQ/MQPRIO slaves, or non |
| 66 | * multiqueue device. |
| 67 | */ |
| 68 | #define TCQ_F_WARN_NONWC (1 << 16) |
| 69 | #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */ |
| 70 | #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy : |
| 71 | * qdisc_tree_decrease_qlen() should stop. |
| 72 | */ |
| 73 | #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */ |
| 74 | #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */ |
| 75 | #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */ |
| 76 | u32 limit; |
| 77 | const struct Qdisc_ops *ops; |
| 78 | struct qdisc_size_table __rcu *stab; |
| 79 | struct hlist_node hash; |
| 80 | u32 handle; |
| 81 | u32 parent; |
| 82 | |
| 83 | struct netdev_queue *dev_queue; |
| 84 | |
| 85 | struct net_rate_estimator __rcu *rate_est; |
| 86 | struct gnet_stats_basic_cpu __percpu *cpu_bstats; |
| 87 | struct gnet_stats_queue __percpu *cpu_qstats; |
| 88 | |
| 89 | /* |
| 90 | * For performance sake on SMP, we put highly modified fields at the end |
| 91 | */ |
| 92 | struct sk_buff_head gso_skb ____cacheline_aligned_in_smp; |
| 93 | struct qdisc_skb_head q; |
| 94 | struct gnet_stats_basic_packed bstats; |
| 95 | seqcount_t running; |
| 96 | struct gnet_stats_queue qstats; |
| 97 | unsigned long state; |
| 98 | struct Qdisc *next_sched; |
| 99 | struct sk_buff_head skb_bad_txq; |
| 100 | int padded; |
| 101 | refcount_t refcnt; |
| 102 | |
| 103 | spinlock_t busylock ____cacheline_aligned_in_smp; |
| 104 | }; |
| 105 | |
| 106 | static inline void qdisc_refcount_inc(struct Qdisc *qdisc) |
| 107 | { |
| 108 | if (qdisc->flags & TCQ_F_BUILTIN) |
| 109 | return; |
| 110 | refcount_inc(&qdisc->refcnt); |
| 111 | } |
| 112 | |
| 113 | static inline bool qdisc_is_running(const struct Qdisc *qdisc) |
| 114 | { |
| 115 | return (raw_read_seqcount(&qdisc->running) & 1) ? true : false; |
| 116 | } |
| 117 | |
| 118 | static inline bool qdisc_run_begin(struct Qdisc *qdisc) |
| 119 | { |
| 120 | if (qdisc_is_running(qdisc)) |
| 121 | return false; |
| 122 | /* Variant of write_seqcount_begin() telling lockdep a trylock |
| 123 | * was attempted. |
| 124 | */ |
| 125 | raw_write_seqcount_begin(&qdisc->running); |
| 126 | seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_); |
| 127 | return true; |
| 128 | } |
| 129 | |
| 130 | static inline void qdisc_run_end(struct Qdisc *qdisc) |
| 131 | { |
| 132 | write_seqcount_end(&qdisc->running); |
| 133 | } |
| 134 | |
| 135 | static inline bool qdisc_may_bulk(const struct Qdisc *qdisc) |
| 136 | { |
| 137 | return qdisc->flags & TCQ_F_ONETXQUEUE; |
| 138 | } |
| 139 | |
| 140 | static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq) |
| 141 | { |
| 142 | #ifdef CONFIG_BQL |
| 143 | /* Non-BQL migrated drivers will return 0, too. */ |
| 144 | return dql_avail(&txq->dql); |
| 145 | #else |
| 146 | return 0; |
| 147 | #endif |
| 148 | } |
| 149 | |
| 150 | struct Qdisc_class_ops { |
| 151 | /* Child qdisc manipulation */ |
| 152 | struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); |
| 153 | int (*graft)(struct Qdisc *, unsigned long cl, |
| 154 | struct Qdisc *, struct Qdisc **, |
| 155 | struct netlink_ext_ack *extack); |
| 156 | struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); |
| 157 | void (*qlen_notify)(struct Qdisc *, unsigned long); |
| 158 | |
| 159 | /* Class manipulation routines */ |
| 160 | unsigned long (*find)(struct Qdisc *, u32 classid); |
| 161 | int (*change)(struct Qdisc *, u32, u32, |
| 162 | struct nlattr **, unsigned long *, |
| 163 | struct netlink_ext_ack *); |
| 164 | int (*delete)(struct Qdisc *, unsigned long); |
| 165 | void (*walk)(struct Qdisc *, struct qdisc_walker * arg); |
| 166 | |
| 167 | /* Filter manipulation */ |
| 168 | struct tcf_block * (*tcf_block)(struct Qdisc *sch, |
| 169 | unsigned long arg, |
| 170 | struct netlink_ext_ack *extack); |
| 171 | unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, |
| 172 | u32 classid); |
| 173 | void (*unbind_tcf)(struct Qdisc *, unsigned long); |
| 174 | |
| 175 | /* rtnetlink specific */ |
| 176 | int (*dump)(struct Qdisc *, unsigned long, |
| 177 | struct sk_buff *skb, struct tcmsg*); |
| 178 | int (*dump_stats)(struct Qdisc *, unsigned long, |
| 179 | struct gnet_dump *); |
| 180 | }; |
| 181 | |
| 182 | struct Qdisc_ops { |
| 183 | struct Qdisc_ops *next; |
| 184 | const struct Qdisc_class_ops *cl_ops; |
| 185 | char id[IFNAMSIZ]; |
| 186 | int priv_size; |
| 187 | unsigned int static_flags; |
| 188 | |
| 189 | int (*enqueue)(struct sk_buff *skb, |
| 190 | struct Qdisc *sch, |
| 191 | struct sk_buff **to_free); |
| 192 | struct sk_buff * (*dequeue)(struct Qdisc *); |
| 193 | struct sk_buff * (*peek)(struct Qdisc *); |
| 194 | |
| 195 | int (*init)(struct Qdisc *sch, struct nlattr *arg, |
| 196 | struct netlink_ext_ack *extack); |
| 197 | void (*reset)(struct Qdisc *); |
| 198 | void (*destroy)(struct Qdisc *); |
| 199 | int (*change)(struct Qdisc *sch, |
| 200 | struct nlattr *arg, |
| 201 | struct netlink_ext_ack *extack); |
| 202 | void (*attach)(struct Qdisc *sch); |
| 203 | int (*change_tx_queue_len)(struct Qdisc *, unsigned int); |
| 204 | |
| 205 | int (*dump)(struct Qdisc *, struct sk_buff *); |
| 206 | int (*dump_stats)(struct Qdisc *, struct gnet_dump *); |
| 207 | |
| 208 | void (*ingress_block_set)(struct Qdisc *sch, |
| 209 | u32 block_index); |
| 210 | void (*egress_block_set)(struct Qdisc *sch, |
| 211 | u32 block_index); |
| 212 | u32 (*ingress_block_get)(struct Qdisc *sch); |
| 213 | u32 (*egress_block_get)(struct Qdisc *sch); |
| 214 | |
| 215 | struct module *owner; |
| 216 | }; |
| 217 | |
| 218 | |
| 219 | struct tcf_result { |
| 220 | union { |
| 221 | struct { |
| 222 | unsigned long class; |
| 223 | u32 classid; |
| 224 | }; |
| 225 | const struct tcf_proto *goto_tp; |
| 226 | }; |
| 227 | }; |
| 228 | |
| 229 | struct tcf_proto_ops { |
| 230 | struct list_head head; |
| 231 | char kind[IFNAMSIZ]; |
| 232 | |
| 233 | int (*classify)(struct sk_buff *, |
| 234 | const struct tcf_proto *, |
| 235 | struct tcf_result *); |
| 236 | int (*init)(struct tcf_proto*); |
| 237 | void (*destroy)(struct tcf_proto *tp, |
| 238 | struct netlink_ext_ack *extack); |
| 239 | |
| 240 | void* (*get)(struct tcf_proto*, u32 handle); |
| 241 | int (*change)(struct net *net, struct sk_buff *, |
| 242 | struct tcf_proto*, unsigned long, |
| 243 | u32 handle, struct nlattr **, |
| 244 | void **, bool, |
| 245 | struct netlink_ext_ack *); |
| 246 | int (*delete)(struct tcf_proto *tp, void *arg, |
| 247 | bool *last, |
| 248 | struct netlink_ext_ack *); |
| 249 | void (*walk)(struct tcf_proto*, struct tcf_walker *arg); |
| 250 | void (*bind_class)(void *, u32, unsigned long); |
| 251 | |
| 252 | /* rtnetlink specific */ |
| 253 | int (*dump)(struct net*, struct tcf_proto*, void *, |
| 254 | struct sk_buff *skb, struct tcmsg*); |
| 255 | |
| 256 | struct module *owner; |
| 257 | }; |
| 258 | |
| 259 | struct tcf_proto { |
| 260 | /* Fast access part */ |
| 261 | struct tcf_proto __rcu *next; |
| 262 | void __rcu *root; |
| 263 | int (*classify)(struct sk_buff *, |
| 264 | const struct tcf_proto *, |
| 265 | struct tcf_result *); |
| 266 | __be16 protocol; |
| 267 | |
| 268 | /* All the rest */ |
| 269 | u32 prio; |
| 270 | void *data; |
| 271 | const struct tcf_proto_ops *ops; |
| 272 | struct tcf_chain *chain; |
| 273 | struct rcu_head rcu; |
| 274 | }; |
| 275 | |
| 276 | struct qdisc_skb_cb { |
| 277 | unsigned int pkt_len; |
| 278 | u16 slave_dev_queue_mapping; |
| 279 | u16 tc_classid; |
| 280 | #define QDISC_CB_PRIV_LEN 20 |
| 281 | unsigned char data[QDISC_CB_PRIV_LEN]; |
| 282 | }; |
| 283 | |
| 284 | typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv); |
| 285 | |
| 286 | struct tcf_chain { |
| 287 | struct tcf_proto __rcu *filter_chain; |
| 288 | struct list_head filter_chain_list; |
| 289 | struct list_head list; |
| 290 | struct tcf_block *block; |
| 291 | u32 index; /* chain index */ |
| 292 | unsigned int refcnt; |
| 293 | }; |
| 294 | |
| 295 | struct tcf_block { |
| 296 | struct list_head chain_list; |
| 297 | u32 index; /* block index for shared blocks */ |
| 298 | unsigned int refcnt; |
| 299 | struct net *net; |
| 300 | struct Qdisc *q; |
| 301 | struct list_head cb_list; |
| 302 | struct list_head owner_list; |
| 303 | bool keep_dst; |
| 304 | unsigned int offloadcnt; /* Number of oddloaded filters */ |
| 305 | unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */ |
| 306 | }; |
| 307 | |
| 308 | static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags) |
| 309 | { |
| 310 | if (*flags & TCA_CLS_FLAGS_IN_HW) |
| 311 | return; |
| 312 | *flags |= TCA_CLS_FLAGS_IN_HW; |
| 313 | block->offloadcnt++; |
| 314 | } |
| 315 | |
| 316 | static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags) |
| 317 | { |
| 318 | if (!(*flags & TCA_CLS_FLAGS_IN_HW)) |
| 319 | return; |
| 320 | *flags &= ~TCA_CLS_FLAGS_IN_HW; |
| 321 | block->offloadcnt--; |
| 322 | } |
| 323 | |
| 324 | static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) |
| 325 | { |
| 326 | struct qdisc_skb_cb *qcb; |
| 327 | |
| 328 | BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz); |
| 329 | BUILD_BUG_ON(sizeof(qcb->data) < sz); |
| 330 | } |
| 331 | |
| 332 | static inline int qdisc_qlen_cpu(const struct Qdisc *q) |
| 333 | { |
| 334 | return this_cpu_ptr(q->cpu_qstats)->qlen; |
| 335 | } |
| 336 | |
| 337 | static inline int qdisc_qlen(const struct Qdisc *q) |
| 338 | { |
| 339 | return q->q.qlen; |
| 340 | } |
| 341 | |
| 342 | static inline int qdisc_qlen_sum(const struct Qdisc *q) |
| 343 | { |
| 344 | __u32 qlen = 0; |
| 345 | int i; |
| 346 | |
| 347 | if (q->flags & TCQ_F_NOLOCK) { |
| 348 | for_each_possible_cpu(i) |
| 349 | qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen; |
| 350 | } else { |
| 351 | qlen = q->q.qlen; |
| 352 | } |
| 353 | |
| 354 | return qlen; |
| 355 | } |
| 356 | |
| 357 | static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) |
| 358 | { |
| 359 | return (struct qdisc_skb_cb *)skb->cb; |
| 360 | } |
| 361 | |
| 362 | static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) |
| 363 | { |
| 364 | return &qdisc->q.lock; |
| 365 | } |
| 366 | |
| 367 | static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) |
| 368 | { |
| 369 | struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc); |
| 370 | |
| 371 | return q; |
| 372 | } |
| 373 | |
| 374 | static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) |
| 375 | { |
| 376 | return qdisc->dev_queue->qdisc_sleeping; |
| 377 | } |
| 378 | |
| 379 | /* The qdisc root lock is a mechanism by which to top level |
| 380 | * of a qdisc tree can be locked from any qdisc node in the |
| 381 | * forest. This allows changing the configuration of some |
| 382 | * aspect of the qdisc tree while blocking out asynchronous |
| 383 | * qdisc access in the packet processing paths. |
| 384 | * |
| 385 | * It is only legal to do this when the root will not change |
| 386 | * on us. Otherwise we'll potentially lock the wrong qdisc |
| 387 | * root. This is enforced by holding the RTNL semaphore, which |
| 388 | * all users of this lock accessor must do. |
| 389 | */ |
| 390 | static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) |
| 391 | { |
| 392 | struct Qdisc *root = qdisc_root(qdisc); |
| 393 | |
| 394 | ASSERT_RTNL(); |
| 395 | return qdisc_lock(root); |
| 396 | } |
| 397 | |
| 398 | static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) |
| 399 | { |
| 400 | struct Qdisc *root = qdisc_root_sleeping(qdisc); |
| 401 | |
| 402 | ASSERT_RTNL(); |
| 403 | return qdisc_lock(root); |
| 404 | } |
| 405 | |
| 406 | static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc) |
| 407 | { |
| 408 | struct Qdisc *root = qdisc_root_sleeping(qdisc); |
| 409 | |
| 410 | ASSERT_RTNL(); |
| 411 | return &root->running; |
| 412 | } |
| 413 | |
| 414 | static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) |
| 415 | { |
| 416 | return qdisc->dev_queue->dev; |
| 417 | } |
| 418 | |
| 419 | static inline void sch_tree_lock(const struct Qdisc *q) |
| 420 | { |
| 421 | spin_lock_bh(qdisc_root_sleeping_lock(q)); |
| 422 | } |
| 423 | |
| 424 | static inline void sch_tree_unlock(const struct Qdisc *q) |
| 425 | { |
| 426 | spin_unlock_bh(qdisc_root_sleeping_lock(q)); |
| 427 | } |
| 428 | |
| 429 | extern struct Qdisc noop_qdisc; |
| 430 | extern struct Qdisc_ops noop_qdisc_ops; |
| 431 | extern struct Qdisc_ops pfifo_fast_ops; |
| 432 | extern struct Qdisc_ops mq_qdisc_ops; |
| 433 | extern struct Qdisc_ops noqueue_qdisc_ops; |
| 434 | extern const struct Qdisc_ops *default_qdisc_ops; |
| 435 | static inline const struct Qdisc_ops * |
| 436 | get_default_qdisc_ops(const struct net_device *dev, int ntx) |
| 437 | { |
| 438 | return ntx < dev->real_num_tx_queues ? |
| 439 | default_qdisc_ops : &pfifo_fast_ops; |
| 440 | } |
| 441 | |
| 442 | struct Qdisc_class_common { |
| 443 | u32 classid; |
| 444 | struct hlist_node hnode; |
| 445 | }; |
| 446 | |
| 447 | struct Qdisc_class_hash { |
| 448 | struct hlist_head *hash; |
| 449 | unsigned int hashsize; |
| 450 | unsigned int hashmask; |
| 451 | unsigned int hashelems; |
| 452 | }; |
| 453 | |
| 454 | static inline unsigned int qdisc_class_hash(u32 id, u32 mask) |
| 455 | { |
| 456 | id ^= id >> 8; |
| 457 | id ^= id >> 4; |
| 458 | return id & mask; |
| 459 | } |
| 460 | |
| 461 | static inline struct Qdisc_class_common * |
| 462 | qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) |
| 463 | { |
| 464 | struct Qdisc_class_common *cl; |
| 465 | unsigned int h; |
| 466 | |
| 467 | if (!id) |
| 468 | return NULL; |
| 469 | |
| 470 | h = qdisc_class_hash(id, hash->hashmask); |
| 471 | hlist_for_each_entry(cl, &hash->hash[h], hnode) { |
| 472 | if (cl->classid == id) |
| 473 | return cl; |
| 474 | } |
| 475 | return NULL; |
| 476 | } |
| 477 | |
| 478 | static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid) |
| 479 | { |
| 480 | u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY; |
| 481 | |
| 482 | return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL; |
| 483 | } |
| 484 | |
| 485 | int qdisc_class_hash_init(struct Qdisc_class_hash *); |
| 486 | void qdisc_class_hash_insert(struct Qdisc_class_hash *, |
| 487 | struct Qdisc_class_common *); |
| 488 | void qdisc_class_hash_remove(struct Qdisc_class_hash *, |
| 489 | struct Qdisc_class_common *); |
| 490 | void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); |
| 491 | void qdisc_class_hash_destroy(struct Qdisc_class_hash *); |
| 492 | |
| 493 | int dev_qdisc_change_tx_queue_len(struct net_device *dev); |
| 494 | void dev_init_scheduler(struct net_device *dev); |
| 495 | void dev_shutdown(struct net_device *dev); |
| 496 | void dev_activate(struct net_device *dev); |
| 497 | void dev_deactivate(struct net_device *dev); |
| 498 | void dev_deactivate_many(struct list_head *head); |
| 499 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, |
| 500 | struct Qdisc *qdisc); |
| 501 | void qdisc_reset(struct Qdisc *qdisc); |
| 502 | void qdisc_destroy(struct Qdisc *qdisc); |
| 503 | void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n, |
| 504 | unsigned int len); |
| 505 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
| 506 | const struct Qdisc_ops *ops, |
| 507 | struct netlink_ext_ack *extack); |
| 508 | void qdisc_free(struct Qdisc *qdisc); |
| 509 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
| 510 | const struct Qdisc_ops *ops, u32 parentid, |
| 511 | struct netlink_ext_ack *extack); |
| 512 | void __qdisc_calculate_pkt_len(struct sk_buff *skb, |
| 513 | const struct qdisc_size_table *stab); |
| 514 | int skb_do_redirect(struct sk_buff *); |
| 515 | |
| 516 | static inline void skb_reset_tc(struct sk_buff *skb) |
| 517 | { |
| 518 | #ifdef CONFIG_NET_CLS_ACT |
| 519 | skb->tc_redirected = 0; |
| 520 | #endif |
| 521 | } |
| 522 | |
| 523 | static inline bool skb_at_tc_ingress(const struct sk_buff *skb) |
| 524 | { |
| 525 | #ifdef CONFIG_NET_CLS_ACT |
| 526 | return skb->tc_at_ingress; |
| 527 | #else |
| 528 | return false; |
| 529 | #endif |
| 530 | } |
| 531 | |
| 532 | static inline bool skb_skip_tc_classify(struct sk_buff *skb) |
| 533 | { |
| 534 | #ifdef CONFIG_NET_CLS_ACT |
| 535 | if (skb->tc_skip_classify) { |
| 536 | skb->tc_skip_classify = 0; |
| 537 | return true; |
| 538 | } |
| 539 | #endif |
| 540 | return false; |
| 541 | } |
| 542 | |
| 543 | /* Reset all TX qdiscs greater then index of a device. */ |
| 544 | static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) |
| 545 | { |
| 546 | struct Qdisc *qdisc; |
| 547 | |
| 548 | for (; i < dev->num_tx_queues; i++) { |
| 549 | qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc); |
| 550 | if (qdisc) { |
| 551 | spin_lock_bh(qdisc_lock(qdisc)); |
| 552 | qdisc_reset(qdisc); |
| 553 | spin_unlock_bh(qdisc_lock(qdisc)); |
| 554 | } |
| 555 | } |
| 556 | } |
| 557 | |
| 558 | static inline void qdisc_reset_all_tx(struct net_device *dev) |
| 559 | { |
| 560 | qdisc_reset_all_tx_gt(dev, 0); |
| 561 | } |
| 562 | |
| 563 | /* Are all TX queues of the device empty? */ |
| 564 | static inline bool qdisc_all_tx_empty(const struct net_device *dev) |
| 565 | { |
| 566 | unsigned int i; |
| 567 | |
| 568 | rcu_read_lock(); |
| 569 | for (i = 0; i < dev->num_tx_queues; i++) { |
| 570 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); |
| 571 | const struct Qdisc *q = rcu_dereference(txq->qdisc); |
| 572 | |
| 573 | if (q->q.qlen) { |
| 574 | rcu_read_unlock(); |
| 575 | return false; |
| 576 | } |
| 577 | } |
| 578 | rcu_read_unlock(); |
| 579 | return true; |
| 580 | } |
| 581 | |
| 582 | /* Are any of the TX qdiscs changing? */ |
| 583 | static inline bool qdisc_tx_changing(const struct net_device *dev) |
| 584 | { |
| 585 | unsigned int i; |
| 586 | |
| 587 | for (i = 0; i < dev->num_tx_queues; i++) { |
| 588 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); |
| 589 | if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping) |
| 590 | return true; |
| 591 | } |
| 592 | return false; |
| 593 | } |
| 594 | |
| 595 | /* Is the device using the noop qdisc on all queues? */ |
| 596 | static inline bool qdisc_tx_is_noop(const struct net_device *dev) |
| 597 | { |
| 598 | unsigned int i; |
| 599 | |
| 600 | for (i = 0; i < dev->num_tx_queues; i++) { |
| 601 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); |
| 602 | if (rcu_access_pointer(txq->qdisc) != &noop_qdisc) |
| 603 | return false; |
| 604 | } |
| 605 | return true; |
| 606 | } |
| 607 | |
| 608 | static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) |
| 609 | { |
| 610 | return qdisc_skb_cb(skb)->pkt_len; |
| 611 | } |
| 612 | |
| 613 | /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ |
| 614 | enum net_xmit_qdisc_t { |
| 615 | __NET_XMIT_STOLEN = 0x00010000, |
| 616 | __NET_XMIT_BYPASS = 0x00020000, |
| 617 | }; |
| 618 | |
| 619 | #ifdef CONFIG_NET_CLS_ACT |
| 620 | #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) |
| 621 | #else |
| 622 | #define net_xmit_drop_count(e) (1) |
| 623 | #endif |
| 624 | |
| 625 | static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, |
| 626 | const struct Qdisc *sch) |
| 627 | { |
| 628 | #ifdef CONFIG_NET_SCHED |
| 629 | struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); |
| 630 | |
| 631 | if (stab) |
| 632 | __qdisc_calculate_pkt_len(skb, stab); |
| 633 | #endif |
| 634 | } |
| 635 | |
| 636 | static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
| 637 | struct sk_buff **to_free) |
| 638 | { |
| 639 | qdisc_calculate_pkt_len(skb, sch); |
| 640 | return sch->enqueue(skb, sch, to_free); |
| 641 | } |
| 642 | |
| 643 | static inline bool qdisc_is_percpu_stats(const struct Qdisc *q) |
| 644 | { |
| 645 | return q->flags & TCQ_F_CPUSTATS; |
| 646 | } |
| 647 | |
| 648 | static inline void _bstats_update(struct gnet_stats_basic_packed *bstats, |
| 649 | __u64 bytes, __u32 packets) |
| 650 | { |
| 651 | bstats->bytes += bytes; |
| 652 | bstats->packets += packets; |
| 653 | } |
| 654 | |
| 655 | static inline void bstats_update(struct gnet_stats_basic_packed *bstats, |
| 656 | const struct sk_buff *skb) |
| 657 | { |
| 658 | _bstats_update(bstats, |
| 659 | qdisc_pkt_len(skb), |
| 660 | skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1); |
| 661 | } |
| 662 | |
| 663 | static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, |
| 664 | __u64 bytes, __u32 packets) |
| 665 | { |
| 666 | u64_stats_update_begin(&bstats->syncp); |
| 667 | _bstats_update(&bstats->bstats, bytes, packets); |
| 668 | u64_stats_update_end(&bstats->syncp); |
| 669 | } |
| 670 | |
| 671 | static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats, |
| 672 | const struct sk_buff *skb) |
| 673 | { |
| 674 | u64_stats_update_begin(&bstats->syncp); |
| 675 | bstats_update(&bstats->bstats, skb); |
| 676 | u64_stats_update_end(&bstats->syncp); |
| 677 | } |
| 678 | |
| 679 | static inline void qdisc_bstats_cpu_update(struct Qdisc *sch, |
| 680 | const struct sk_buff *skb) |
| 681 | { |
| 682 | bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb); |
| 683 | } |
| 684 | |
| 685 | static inline void qdisc_bstats_update(struct Qdisc *sch, |
| 686 | const struct sk_buff *skb) |
| 687 | { |
| 688 | bstats_update(&sch->bstats, skb); |
| 689 | } |
| 690 | |
| 691 | static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch, |
| 692 | const struct sk_buff *skb) |
| 693 | { |
| 694 | sch->qstats.backlog -= qdisc_pkt_len(skb); |
| 695 | } |
| 696 | |
| 697 | static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch, |
| 698 | const struct sk_buff *skb) |
| 699 | { |
| 700 | this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); |
| 701 | } |
| 702 | |
| 703 | static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch, |
| 704 | const struct sk_buff *skb) |
| 705 | { |
| 706 | sch->qstats.backlog += qdisc_pkt_len(skb); |
| 707 | } |
| 708 | |
| 709 | static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch, |
| 710 | const struct sk_buff *skb) |
| 711 | { |
| 712 | this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb)); |
| 713 | } |
| 714 | |
| 715 | static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch) |
| 716 | { |
| 717 | this_cpu_inc(sch->cpu_qstats->qlen); |
| 718 | } |
| 719 | |
| 720 | static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch) |
| 721 | { |
| 722 | this_cpu_dec(sch->cpu_qstats->qlen); |
| 723 | } |
| 724 | |
| 725 | static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch) |
| 726 | { |
| 727 | this_cpu_inc(sch->cpu_qstats->requeues); |
| 728 | } |
| 729 | |
| 730 | static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count) |
| 731 | { |
| 732 | sch->qstats.drops += count; |
| 733 | } |
| 734 | |
| 735 | static inline void qstats_drop_inc(struct gnet_stats_queue *qstats) |
| 736 | { |
| 737 | qstats->drops++; |
| 738 | } |
| 739 | |
| 740 | static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats) |
| 741 | { |
| 742 | qstats->overlimits++; |
| 743 | } |
| 744 | |
| 745 | static inline void qdisc_qstats_drop(struct Qdisc *sch) |
| 746 | { |
| 747 | qstats_drop_inc(&sch->qstats); |
| 748 | } |
| 749 | |
| 750 | static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch) |
| 751 | { |
| 752 | this_cpu_inc(sch->cpu_qstats->drops); |
| 753 | } |
| 754 | |
| 755 | static inline void qdisc_qstats_overlimit(struct Qdisc *sch) |
| 756 | { |
| 757 | sch->qstats.overlimits++; |
| 758 | } |
| 759 | |
| 760 | static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh) |
| 761 | { |
| 762 | qh->head = NULL; |
| 763 | qh->tail = NULL; |
| 764 | qh->qlen = 0; |
| 765 | } |
| 766 | |
| 767 | static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch, |
| 768 | struct qdisc_skb_head *qh) |
| 769 | { |
| 770 | struct sk_buff *last = qh->tail; |
| 771 | |
| 772 | if (last) { |
| 773 | skb->next = NULL; |
| 774 | last->next = skb; |
| 775 | qh->tail = skb; |
| 776 | } else { |
| 777 | qh->tail = skb; |
| 778 | qh->head = skb; |
| 779 | } |
| 780 | qh->qlen++; |
| 781 | qdisc_qstats_backlog_inc(sch, skb); |
| 782 | |
| 783 | return NET_XMIT_SUCCESS; |
| 784 | } |
| 785 | |
| 786 | static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) |
| 787 | { |
| 788 | return __qdisc_enqueue_tail(skb, sch, &sch->q); |
| 789 | } |
| 790 | |
| 791 | static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh) |
| 792 | { |
| 793 | struct sk_buff *skb = qh->head; |
| 794 | |
| 795 | if (likely(skb != NULL)) { |
| 796 | qh->head = skb->next; |
| 797 | qh->qlen--; |
| 798 | if (qh->head == NULL) |
| 799 | qh->tail = NULL; |
| 800 | skb->next = NULL; |
| 801 | } |
| 802 | |
| 803 | return skb; |
| 804 | } |
| 805 | |
| 806 | static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) |
| 807 | { |
| 808 | struct sk_buff *skb = __qdisc_dequeue_head(&sch->q); |
| 809 | |
| 810 | if (likely(skb != NULL)) { |
| 811 | qdisc_qstats_backlog_dec(sch, skb); |
| 812 | qdisc_bstats_update(sch, skb); |
| 813 | } |
| 814 | |
| 815 | return skb; |
| 816 | } |
| 817 | |
| 818 | /* Instead of calling kfree_skb() while root qdisc lock is held, |
| 819 | * queue the skb for future freeing at end of __dev_xmit_skb() |
| 820 | */ |
| 821 | static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free) |
| 822 | { |
| 823 | skb->next = *to_free; |
| 824 | *to_free = skb; |
| 825 | } |
| 826 | |
| 827 | static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, |
| 828 | struct qdisc_skb_head *qh, |
| 829 | struct sk_buff **to_free) |
| 830 | { |
| 831 | struct sk_buff *skb = __qdisc_dequeue_head(qh); |
| 832 | |
| 833 | if (likely(skb != NULL)) { |
| 834 | unsigned int len = qdisc_pkt_len(skb); |
| 835 | |
| 836 | qdisc_qstats_backlog_dec(sch, skb); |
| 837 | __qdisc_drop(skb, to_free); |
| 838 | return len; |
| 839 | } |
| 840 | |
| 841 | return 0; |
| 842 | } |
| 843 | |
| 844 | static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch, |
| 845 | struct sk_buff **to_free) |
| 846 | { |
| 847 | return __qdisc_queue_drop_head(sch, &sch->q, to_free); |
| 848 | } |
| 849 | |
| 850 | static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) |
| 851 | { |
| 852 | const struct qdisc_skb_head *qh = &sch->q; |
| 853 | |
| 854 | return qh->head; |
| 855 | } |
| 856 | |
| 857 | /* generic pseudo peek method for non-work-conserving qdisc */ |
| 858 | static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) |
| 859 | { |
| 860 | struct sk_buff *skb = skb_peek(&sch->gso_skb); |
| 861 | |
| 862 | /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ |
| 863 | if (!skb) { |
| 864 | skb = sch->dequeue(sch); |
| 865 | |
| 866 | if (skb) { |
| 867 | __skb_queue_head(&sch->gso_skb, skb); |
| 868 | /* it's still part of the queue */ |
| 869 | qdisc_qstats_backlog_inc(sch, skb); |
| 870 | sch->q.qlen++; |
| 871 | } |
| 872 | } |
| 873 | |
| 874 | return skb; |
| 875 | } |
| 876 | |
| 877 | /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ |
| 878 | static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) |
| 879 | { |
| 880 | struct sk_buff *skb = skb_peek(&sch->gso_skb); |
| 881 | |
| 882 | if (skb) { |
| 883 | skb = __skb_dequeue(&sch->gso_skb); |
| 884 | qdisc_qstats_backlog_dec(sch, skb); |
| 885 | sch->q.qlen--; |
| 886 | } else { |
| 887 | skb = sch->dequeue(sch); |
| 888 | } |
| 889 | |
| 890 | return skb; |
| 891 | } |
| 892 | |
| 893 | static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh) |
| 894 | { |
| 895 | /* |
| 896 | * We do not know the backlog in bytes of this list, it |
| 897 | * is up to the caller to correct it |
| 898 | */ |
| 899 | ASSERT_RTNL(); |
| 900 | if (qh->qlen) { |
| 901 | rtnl_kfree_skbs(qh->head, qh->tail); |
| 902 | |
| 903 | qh->head = NULL; |
| 904 | qh->tail = NULL; |
| 905 | qh->qlen = 0; |
| 906 | } |
| 907 | } |
| 908 | |
| 909 | static inline void qdisc_reset_queue(struct Qdisc *sch) |
| 910 | { |
| 911 | __qdisc_reset_queue(&sch->q); |
| 912 | sch->qstats.backlog = 0; |
| 913 | } |
| 914 | |
| 915 | static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new, |
| 916 | struct Qdisc **pold) |
| 917 | { |
| 918 | struct Qdisc *old; |
| 919 | |
| 920 | sch_tree_lock(sch); |
| 921 | old = *pold; |
| 922 | *pold = new; |
| 923 | if (old != NULL) { |
| 924 | unsigned int qlen = old->q.qlen; |
| 925 | unsigned int backlog = old->qstats.backlog; |
| 926 | |
| 927 | qdisc_reset(old); |
| 928 | qdisc_tree_reduce_backlog(old, qlen, backlog); |
| 929 | } |
| 930 | sch_tree_unlock(sch); |
| 931 | |
| 932 | return old; |
| 933 | } |
| 934 | |
| 935 | static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) |
| 936 | { |
| 937 | rtnl_kfree_skbs(skb, skb); |
| 938 | qdisc_qstats_drop(sch); |
| 939 | } |
| 940 | |
| 941 | static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch, |
| 942 | struct sk_buff **to_free) |
| 943 | { |
| 944 | __qdisc_drop(skb, to_free); |
| 945 | qdisc_qstats_cpu_drop(sch); |
| 946 | |
| 947 | return NET_XMIT_DROP; |
| 948 | } |
| 949 | |
| 950 | static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch, |
| 951 | struct sk_buff **to_free) |
| 952 | { |
| 953 | __qdisc_drop(skb, to_free); |
| 954 | qdisc_qstats_drop(sch); |
| 955 | |
| 956 | return NET_XMIT_DROP; |
| 957 | } |
| 958 | |
| 959 | /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how |
| 960 | long it will take to send a packet given its size. |
| 961 | */ |
| 962 | static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) |
| 963 | { |
| 964 | int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; |
| 965 | if (slot < 0) |
| 966 | slot = 0; |
| 967 | slot >>= rtab->rate.cell_log; |
| 968 | if (slot > 255) |
| 969 | return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; |
| 970 | return rtab->data[slot]; |
| 971 | } |
| 972 | |
| 973 | struct psched_ratecfg { |
| 974 | u64 rate_bytes_ps; /* bytes per second */ |
| 975 | u32 mult; |
| 976 | u16 overhead; |
| 977 | u8 linklayer; |
| 978 | u8 shift; |
| 979 | }; |
| 980 | |
| 981 | static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, |
| 982 | unsigned int len) |
| 983 | { |
| 984 | len += r->overhead; |
| 985 | |
| 986 | if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) |
| 987 | return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift; |
| 988 | |
| 989 | return ((u64)len * r->mult) >> r->shift; |
| 990 | } |
| 991 | |
| 992 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
| 993 | const struct tc_ratespec *conf, |
| 994 | u64 rate64); |
| 995 | |
| 996 | static inline void psched_ratecfg_getrate(struct tc_ratespec *res, |
| 997 | const struct psched_ratecfg *r) |
| 998 | { |
| 999 | memset(res, 0, sizeof(*res)); |
| 1000 | |
| 1001 | /* legacy struct tc_ratespec has a 32bit @rate field |
| 1002 | * Qdisc using 64bit rate should add new attributes |
| 1003 | * in order to maintain compatibility. |
| 1004 | */ |
| 1005 | res->rate = min_t(u64, r->rate_bytes_ps, ~0U); |
| 1006 | |
| 1007 | res->overhead = r->overhead; |
| 1008 | res->linklayer = (r->linklayer & TC_LINKLAYER_MASK); |
| 1009 | } |
| 1010 | |
| 1011 | /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc. |
| 1012 | * The fast path only needs to access filter list and to update stats |
| 1013 | */ |
| 1014 | struct mini_Qdisc { |
| 1015 | struct tcf_proto *filter_list; |
| 1016 | struct gnet_stats_basic_cpu __percpu *cpu_bstats; |
| 1017 | struct gnet_stats_queue __percpu *cpu_qstats; |
| 1018 | struct rcu_head rcu; |
| 1019 | }; |
| 1020 | |
| 1021 | static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq, |
| 1022 | const struct sk_buff *skb) |
| 1023 | { |
| 1024 | bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb); |
| 1025 | } |
| 1026 | |
| 1027 | static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq) |
| 1028 | { |
| 1029 | this_cpu_inc(miniq->cpu_qstats->drops); |
| 1030 | } |
| 1031 | |
| 1032 | struct mini_Qdisc_pair { |
| 1033 | struct mini_Qdisc miniq1; |
| 1034 | struct mini_Qdisc miniq2; |
| 1035 | struct mini_Qdisc __rcu **p_miniq; |
| 1036 | }; |
| 1037 | |
| 1038 | void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, |
| 1039 | struct tcf_proto *tp_head); |
| 1040 | void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, |
| 1041 | struct mini_Qdisc __rcu **p_miniq); |
| 1042 | |
| 1043 | #endif |