Linux Audio

Check our new training course

Loading...
v4.17
   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
  18struct Qdisc_ops;
  19struct qdisc_walker;
  20struct tcf_walker;
  21struct module;
 
  22
  23struct qdisc_rate_table {
  24	struct tc_ratespec rate;
  25	u32		data[256];
  26	struct qdisc_rate_table *next;
  27	int		refcnt;
  28};
  29
  30enum qdisc_state_t {
  31	__QDISC_STATE_SCHED,
  32	__QDISC_STATE_DEACTIVATED,
  33	__QDISC_STATE_RUNNING,
 
  34};
  35
 
 
 
 
 
 
 
 
 
 
 
 
 
  36struct qdisc_size_table {
  37	struct rcu_head		rcu;
  38	struct list_head	list;
  39	struct tc_sizespec	szopts;
  40	int			refcnt;
  41	u16			data[];
  42};
  43
  44/* similar to sk_buff_head, but skb->prev pointer is undefined. */
  45struct qdisc_skb_head {
  46	struct sk_buff	*head;
  47	struct sk_buff	*tail;
  48	__u32		qlen;
  49	spinlock_t	lock;
  50};
  51
  52struct Qdisc {
  53	int 			(*enqueue)(struct sk_buff *skb,
  54					   struct Qdisc *sch,
  55					   struct sk_buff **to_free);
  56	struct sk_buff *	(*dequeue)(struct Qdisc *sch);
  57	unsigned int		flags;
  58#define TCQ_F_BUILTIN		1
  59#define TCQ_F_INGRESS		2
  60#define TCQ_F_CAN_BYPASS	4
  61#define TCQ_F_MQROOT		8
  62#define TCQ_F_ONETXQUEUE	0x10 /* dequeue_skb() can assume all skbs are for
  63				      * q->dev_queue : It can test
  64				      * netif_xmit_frozen_or_stopped() before
  65				      * dequeueing next packet.
  66				      * Its true for MQ/MQPRIO slaves, or non
  67				      * multiqueue device.
  68				      */
  69#define TCQ_F_WARN_NONWC	(1 << 16)
  70#define TCQ_F_CPUSTATS		0x20 /* run using percpu statistics */
  71#define TCQ_F_NOPARENT		0x40 /* root of its hierarchy :
  72				      * qdisc_tree_decrease_qlen() should stop.
  73				      */
  74#define TCQ_F_INVISIBLE		0x80 /* invisible by default in dump */
  75#define TCQ_F_NOLOCK		0x100 /* qdisc does not require locking */
  76#define TCQ_F_OFFLOADED		0x200 /* qdisc is offloaded to HW */
  77	u32			limit;
  78	const struct Qdisc_ops	*ops;
  79	struct qdisc_size_table	__rcu *stab;
  80	struct hlist_node       hash;
  81	u32			handle;
  82	u32			parent;
  83
  84	struct netdev_queue	*dev_queue;
  85
  86	struct net_rate_estimator __rcu *rate_est;
  87	struct gnet_stats_basic_cpu __percpu *cpu_bstats;
  88	struct gnet_stats_queue	__percpu *cpu_qstats;
 
 
  89
  90	/*
  91	 * For performance sake on SMP, we put highly modified fields at the end
  92	 */
  93	struct sk_buff_head	gso_skb ____cacheline_aligned_in_smp;
  94	struct qdisc_skb_head	q;
  95	struct gnet_stats_basic_packed bstats;
  96	seqcount_t		running;
  97	struct gnet_stats_queue	qstats;
  98	unsigned long		state;
 
  99	struct Qdisc            *next_sched;
 100	struct sk_buff_head	skb_bad_txq;
 101	int			padded;
 102	refcount_t		refcnt;
 103
 104	spinlock_t		busylock ____cacheline_aligned_in_smp;
 
 
 
 
 
 
 105};
 106
 107static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
 108{
 109	if (qdisc->flags & TCQ_F_BUILTIN)
 110		return;
 111	refcount_inc(&qdisc->refcnt);
 112}
 113
 114static inline bool qdisc_is_running(const struct Qdisc *qdisc)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 115{
 116	return (raw_read_seqcount(&qdisc->running) & 1) ? true : false;
 
 
 117}
 118
 
 
 
 119static inline bool qdisc_run_begin(struct Qdisc *qdisc)
 120{
 121	if (qdisc_is_running(qdisc))
 122		return false;
 123	/* Variant of write_seqcount_begin() telling lockdep a trylock
 124	 * was attempted.
 125	 */
 126	raw_write_seqcount_begin(&qdisc->running);
 127	seqcount_acquire(&qdisc->running.dep_map, 0, 1, _RET_IP_);
 128	return true;
 
 
 
 
 
 
 
 
 
 
 
 129}
 130
 131static inline void qdisc_run_end(struct Qdisc *qdisc)
 132{
 133	write_seqcount_end(&qdisc->running);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 134}
 135
 136static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
 137{
 138	return qdisc->flags & TCQ_F_ONETXQUEUE;
 139}
 140
 141static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
 142{
 143#ifdef CONFIG_BQL
 144	/* Non-BQL migrated drivers will return 0, too. */
 145	return dql_avail(&txq->dql);
 146#else
 147	return 0;
 148#endif
 149}
 150
 151struct Qdisc_class_ops {
 
 152	/* Child qdisc manipulation */
 153	struct netdev_queue *	(*select_queue)(struct Qdisc *, struct tcmsg *);
 154	int			(*graft)(struct Qdisc *, unsigned long cl,
 155					struct Qdisc *, struct Qdisc **,
 156					struct netlink_ext_ack *extack);
 157	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
 158	void			(*qlen_notify)(struct Qdisc *, unsigned long);
 159
 160	/* Class manipulation routines */
 161	unsigned long		(*find)(struct Qdisc *, u32 classid);
 162	int			(*change)(struct Qdisc *, u32, u32,
 163					struct nlattr **, unsigned long *,
 164					struct netlink_ext_ack *);
 165	int			(*delete)(struct Qdisc *, unsigned long);
 
 166	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
 167
 168	/* Filter manipulation */
 169	struct tcf_block *	(*tcf_block)(struct Qdisc *sch,
 170					     unsigned long arg,
 171					     struct netlink_ext_ack *extack);
 172	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
 173					u32 classid);
 174	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
 175
 176	/* rtnetlink specific */
 177	int			(*dump)(struct Qdisc *, unsigned long,
 178					struct sk_buff *skb, struct tcmsg*);
 179	int			(*dump_stats)(struct Qdisc *, unsigned long,
 180					struct gnet_dump *);
 181};
 182
 
 
 
 
 
 
 
 183struct Qdisc_ops {
 184	struct Qdisc_ops	*next;
 185	const struct Qdisc_class_ops	*cl_ops;
 186	char			id[IFNAMSIZ];
 187	int			priv_size;
 188	unsigned int		static_flags;
 189
 190	int 			(*enqueue)(struct sk_buff *skb,
 191					   struct Qdisc *sch,
 192					   struct sk_buff **to_free);
 193	struct sk_buff *	(*dequeue)(struct Qdisc *);
 194	struct sk_buff *	(*peek)(struct Qdisc *);
 195
 196	int			(*init)(struct Qdisc *sch, struct nlattr *arg,
 197					struct netlink_ext_ack *extack);
 198	void			(*reset)(struct Qdisc *);
 199	void			(*destroy)(struct Qdisc *);
 200	int			(*change)(struct Qdisc *sch,
 201					  struct nlattr *arg,
 202					  struct netlink_ext_ack *extack);
 203	void			(*attach)(struct Qdisc *sch);
 204	int			(*change_tx_queue_len)(struct Qdisc *, unsigned int);
 
 
 205
 206	int			(*dump)(struct Qdisc *, struct sk_buff *);
 207	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
 208
 209	void			(*ingress_block_set)(struct Qdisc *sch,
 210						     u32 block_index);
 211	void			(*egress_block_set)(struct Qdisc *sch,
 212						    u32 block_index);
 213	u32			(*ingress_block_get)(struct Qdisc *sch);
 214	u32			(*egress_block_get)(struct Qdisc *sch);
 215
 216	struct module		*owner;
 217};
 218
 219
 220struct tcf_result {
 221	union {
 222		struct {
 223			unsigned long	class;
 224			u32		classid;
 225		};
 226		const struct tcf_proto *goto_tp;
 
 227	};
 228};
 229
 
 
 230struct tcf_proto_ops {
 231	struct list_head	head;
 232	char			kind[IFNAMSIZ];
 233
 234	int			(*classify)(struct sk_buff *,
 235					    const struct tcf_proto *,
 236					    struct tcf_result *);
 237	int			(*init)(struct tcf_proto*);
 238	void			(*destroy)(struct tcf_proto *tp,
 239					   struct netlink_ext_ack *extack);
 240
 241	void*			(*get)(struct tcf_proto*, u32 handle);
 
 242	int			(*change)(struct net *net, struct sk_buff *,
 243					struct tcf_proto*, unsigned long,
 244					u32 handle, struct nlattr **,
 245					void **, bool,
 246					struct netlink_ext_ack *);
 247	int			(*delete)(struct tcf_proto *tp, void *arg,
 248					  bool *last,
 249					  struct netlink_ext_ack *);
 250	void			(*walk)(struct tcf_proto*, struct tcf_walker *arg);
 251	void			(*bind_class)(void *, u32, unsigned long);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 252
 253	/* rtnetlink specific */
 254	int			(*dump)(struct net*, struct tcf_proto*, void *,
 255					struct sk_buff *skb, struct tcmsg*);
 
 
 
 
 
 
 
 
 256
 257	struct module		*owner;
 
 
 
 
 
 
 
 
 
 258};
 259
 260struct tcf_proto {
 261	/* Fast access part */
 262	struct tcf_proto __rcu	*next;
 263	void __rcu		*root;
 
 
 264	int			(*classify)(struct sk_buff *,
 265					    const struct tcf_proto *,
 266					    struct tcf_result *);
 267	__be16			protocol;
 268
 269	/* All the rest */
 270	u32			prio;
 271	void			*data;
 272	const struct tcf_proto_ops	*ops;
 273	struct tcf_chain	*chain;
 
 
 
 
 
 
 274	struct rcu_head		rcu;
 
 275};
 276
 277struct qdisc_skb_cb {
 278	unsigned int		pkt_len;
 279	u16			slave_dev_queue_mapping;
 280	u16			tc_classid;
 
 
 281#define QDISC_CB_PRIV_LEN 20
 282	unsigned char		data[QDISC_CB_PRIV_LEN];
 283};
 284
 285typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
 286
 287struct tcf_chain {
 
 
 288	struct tcf_proto __rcu *filter_chain;
 289	struct list_head filter_chain_list;
 290	struct list_head list;
 291	struct tcf_block *block;
 292	u32 index; /* chain index */
 293	unsigned int refcnt;
 
 
 
 
 
 
 294};
 295
 296struct tcf_block {
 
 
 
 
 297	struct list_head chain_list;
 298	u32 index; /* block index for shared blocks */
 299	unsigned int refcnt;
 
 300	struct net *net;
 301	struct Qdisc *q;
 302	struct list_head cb_list;
 
 303	struct list_head owner_list;
 304	bool keep_dst;
 305	unsigned int offloadcnt; /* Number of oddloaded filters */
 306	unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
 
 
 
 
 
 
 
 
 307};
 308
 309static inline void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
 310{
 311	if (*flags & TCA_CLS_FLAGS_IN_HW)
 312		return;
 313	*flags |= TCA_CLS_FLAGS_IN_HW;
 314	block->offloadcnt++;
 315}
 316
 317static inline void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
 318{
 319	if (!(*flags & TCA_CLS_FLAGS_IN_HW))
 320		return;
 321	*flags &= ~TCA_CLS_FLAGS_IN_HW;
 322	block->offloadcnt--;
 323}
 324
 
 
 
 
 
 
 325static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
 326{
 327	struct qdisc_skb_cb *qcb;
 328
 329	BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz);
 330	BUILD_BUG_ON(sizeof(qcb->data) < sz);
 331}
 332
 333static inline int qdisc_qlen_cpu(const struct Qdisc *q)
 334{
 335	return this_cpu_ptr(q->cpu_qstats)->qlen;
 336}
 337
 338static inline int qdisc_qlen(const struct Qdisc *q)
 339{
 340	return q->q.qlen;
 341}
 342
 343static inline int qdisc_qlen_sum(const struct Qdisc *q)
 344{
 345	__u32 qlen = 0;
 346	int i;
 347
 348	if (q->flags & TCQ_F_NOLOCK) {
 349		for_each_possible_cpu(i)
 350			qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
 351	} else {
 352		qlen = q->q.qlen;
 353	}
 354
 355	return qlen;
 356}
 357
 358static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
 359{
 360	return (struct qdisc_skb_cb *)skb->cb;
 361}
 362
 363static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
 364{
 365	return &qdisc->q.lock;
 366}
 367
 368static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
 369{
 370	struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
 371
 372	return q;
 373}
 374
 375static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
 376{
 377	return qdisc->dev_queue->qdisc_sleeping;
 378}
 379
 380/* The qdisc root lock is a mechanism by which to top level
 381 * of a qdisc tree can be locked from any qdisc node in the
 382 * forest.  This allows changing the configuration of some
 383 * aspect of the qdisc tree while blocking out asynchronous
 384 * qdisc access in the packet processing paths.
 385 *
 386 * It is only legal to do this when the root will not change
 387 * on us.  Otherwise we'll potentially lock the wrong qdisc
 388 * root.  This is enforced by holding the RTNL semaphore, which
 389 * all users of this lock accessor must do.
 390 */
 391static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc)
 392{
 393	struct Qdisc *root = qdisc_root(qdisc);
 394
 395	ASSERT_RTNL();
 396	return qdisc_lock(root);
 397}
 398
 399static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
 400{
 401	struct Qdisc *root = qdisc_root_sleeping(qdisc);
 402
 403	ASSERT_RTNL();
 404	return qdisc_lock(root);
 405}
 406
 407static inline seqcount_t *qdisc_root_sleeping_running(const struct Qdisc *qdisc)
 408{
 409	struct Qdisc *root = qdisc_root_sleeping(qdisc);
 410
 411	ASSERT_RTNL();
 412	return &root->running;
 413}
 414
 415static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
 416{
 417	return qdisc->dev_queue->dev;
 418}
 419
 420static inline void sch_tree_lock(const struct Qdisc *q)
 421{
 422	spin_lock_bh(qdisc_root_sleeping_lock(q));
 
 
 
 423}
 424
 425static inline void sch_tree_unlock(const struct Qdisc *q)
 426{
 427	spin_unlock_bh(qdisc_root_sleeping_lock(q));
 
 
 
 428}
 429
 430extern struct Qdisc noop_qdisc;
 431extern struct Qdisc_ops noop_qdisc_ops;
 432extern struct Qdisc_ops pfifo_fast_ops;
 433extern struct Qdisc_ops mq_qdisc_ops;
 434extern struct Qdisc_ops noqueue_qdisc_ops;
 435extern const struct Qdisc_ops *default_qdisc_ops;
 436static inline const struct Qdisc_ops *
 437get_default_qdisc_ops(const struct net_device *dev, int ntx)
 438{
 439	return ntx < dev->real_num_tx_queues ?
 440			default_qdisc_ops : &pfifo_fast_ops;
 441}
 442
 443struct Qdisc_class_common {
 444	u32			classid;
 445	struct hlist_node	hnode;
 446};
 447
 448struct Qdisc_class_hash {
 449	struct hlist_head	*hash;
 450	unsigned int		hashsize;
 451	unsigned int		hashmask;
 452	unsigned int		hashelems;
 453};
 454
 455static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
 456{
 457	id ^= id >> 8;
 458	id ^= id >> 4;
 459	return id & mask;
 460}
 461
 462static inline struct Qdisc_class_common *
 463qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
 464{
 465	struct Qdisc_class_common *cl;
 466	unsigned int h;
 467
 468	if (!id)
 469		return NULL;
 470
 471	h = qdisc_class_hash(id, hash->hashmask);
 472	hlist_for_each_entry(cl, &hash->hash[h], hnode) {
 473		if (cl->classid == id)
 474			return cl;
 475	}
 476	return NULL;
 477}
 478
 479static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
 480{
 481	u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
 482
 483	return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
 484}
 485
 486int qdisc_class_hash_init(struct Qdisc_class_hash *);
 487void qdisc_class_hash_insert(struct Qdisc_class_hash *,
 488			     struct Qdisc_class_common *);
 489void qdisc_class_hash_remove(struct Qdisc_class_hash *,
 490			     struct Qdisc_class_common *);
 491void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
 492void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
 493
 494int dev_qdisc_change_tx_queue_len(struct net_device *dev);
 
 
 495void dev_init_scheduler(struct net_device *dev);
 496void dev_shutdown(struct net_device *dev);
 497void dev_activate(struct net_device *dev);
 498void dev_deactivate(struct net_device *dev);
 499void dev_deactivate_many(struct list_head *head);
 500struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
 501			      struct Qdisc *qdisc);
 502void qdisc_reset(struct Qdisc *qdisc);
 503void qdisc_destroy(struct Qdisc *qdisc);
 504void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, unsigned int n,
 505			       unsigned int len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 506struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
 507			  const struct Qdisc_ops *ops,
 508			  struct netlink_ext_ack *extack);
 509void qdisc_free(struct Qdisc *qdisc);
 510struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
 511				const struct Qdisc_ops *ops, u32 parentid,
 512				struct netlink_ext_ack *extack);
 513void __qdisc_calculate_pkt_len(struct sk_buff *skb,
 514			       const struct qdisc_size_table *stab);
 515int skb_do_redirect(struct sk_buff *);
 516
 517static inline void skb_reset_tc(struct sk_buff *skb)
 518{
 519#ifdef CONFIG_NET_CLS_ACT
 520	skb->tc_redirected = 0;
 521#endif
 522}
 523
 524static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
 525{
 526#ifdef CONFIG_NET_CLS_ACT
 527	return skb->tc_at_ingress;
 528#else
 529	return false;
 530#endif
 531}
 532
 533static inline bool skb_skip_tc_classify(struct sk_buff *skb)
 534{
 535#ifdef CONFIG_NET_CLS_ACT
 536	if (skb->tc_skip_classify) {
 537		skb->tc_skip_classify = 0;
 538		return true;
 539	}
 540#endif
 541	return false;
 542}
 543
 544/* Reset all TX qdiscs greater than index of a device.  */
 545static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
 546{
 547	struct Qdisc *qdisc;
 548
 549	for (; i < dev->num_tx_queues; i++) {
 550		qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
 551		if (qdisc) {
 552			spin_lock_bh(qdisc_lock(qdisc));
 553			qdisc_reset(qdisc);
 554			spin_unlock_bh(qdisc_lock(qdisc));
 555		}
 556	}
 557}
 558
 559static inline void qdisc_reset_all_tx(struct net_device *dev)
 560{
 561	qdisc_reset_all_tx_gt(dev, 0);
 562}
 563
 564/* Are all TX queues of the device empty?  */
 565static inline bool qdisc_all_tx_empty(const struct net_device *dev)
 566{
 567	unsigned int i;
 568
 569	rcu_read_lock();
 570	for (i = 0; i < dev->num_tx_queues; i++) {
 571		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 572		const struct Qdisc *q = rcu_dereference(txq->qdisc);
 573
 574		if (q->q.qlen) {
 575			rcu_read_unlock();
 576			return false;
 577		}
 578	}
 579	rcu_read_unlock();
 580	return true;
 581}
 582
 583/* Are any of the TX qdiscs changing?  */
 584static inline bool qdisc_tx_changing(const struct net_device *dev)
 585{
 586	unsigned int i;
 587
 588	for (i = 0; i < dev->num_tx_queues; i++) {
 589		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 590		if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
 591			return true;
 592	}
 593	return false;
 594}
 595
 596/* Is the device using the noop qdisc on all queues?  */
 597static inline bool qdisc_tx_is_noop(const struct net_device *dev)
 598{
 599	unsigned int i;
 600
 601	for (i = 0; i < dev->num_tx_queues; i++) {
 602		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 603		if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
 604			return false;
 605	}
 606	return true;
 607}
 608
 609static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
 610{
 611	return qdisc_skb_cb(skb)->pkt_len;
 612}
 613
 614/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
 615enum net_xmit_qdisc_t {
 616	__NET_XMIT_STOLEN = 0x00010000,
 617	__NET_XMIT_BYPASS = 0x00020000,
 618};
 619
 620#ifdef CONFIG_NET_CLS_ACT
 621#define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
 622#else
 623#define net_xmit_drop_count(e)	(1)
 624#endif
 625
 626static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
 627					   const struct Qdisc *sch)
 628{
 629#ifdef CONFIG_NET_SCHED
 630	struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
 631
 632	if (stab)
 633		__qdisc_calculate_pkt_len(skb, stab);
 634#endif
 635}
 636
 637static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 638				struct sk_buff **to_free)
 639{
 640	qdisc_calculate_pkt_len(skb, sch);
 641	return sch->enqueue(skb, sch, to_free);
 642}
 643
 644static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
 645{
 646	return q->flags & TCQ_F_CPUSTATS;
 647}
 648
 649static inline void _bstats_update(struct gnet_stats_basic_packed *bstats,
 650				  __u64 bytes, __u32 packets)
 651{
 652	bstats->bytes += bytes;
 653	bstats->packets += packets;
 
 
 654}
 655
 656static inline void bstats_update(struct gnet_stats_basic_packed *bstats,
 657				 const struct sk_buff *skb)
 658{
 659	_bstats_update(bstats,
 660		       qdisc_pkt_len(skb),
 661		       skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
 662}
 663
 664static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
 665				      __u64 bytes, __u32 packets)
 666{
 667	u64_stats_update_begin(&bstats->syncp);
 668	_bstats_update(&bstats->bstats, bytes, packets);
 669	u64_stats_update_end(&bstats->syncp);
 670}
 671
 672static inline void bstats_cpu_update(struct gnet_stats_basic_cpu *bstats,
 673				     const struct sk_buff *skb)
 674{
 675	u64_stats_update_begin(&bstats->syncp);
 676	bstats_update(&bstats->bstats, skb);
 677	u64_stats_update_end(&bstats->syncp);
 678}
 679
 680static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
 681					   const struct sk_buff *skb)
 682{
 683	bstats_cpu_update(this_cpu_ptr(sch->cpu_bstats), skb);
 684}
 685
 686static inline void qdisc_bstats_update(struct Qdisc *sch,
 687				       const struct sk_buff *skb)
 688{
 689	bstats_update(&sch->bstats, skb);
 690}
 691
 692static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
 693					    const struct sk_buff *skb)
 694{
 695	sch->qstats.backlog -= qdisc_pkt_len(skb);
 696}
 697
 698static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
 699						const struct sk_buff *skb)
 700{
 701	this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
 702}
 703
 704static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
 705					    const struct sk_buff *skb)
 706{
 707	sch->qstats.backlog += qdisc_pkt_len(skb);
 708}
 709
 710static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
 711						const struct sk_buff *skb)
 712{
 713	this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
 714}
 715
 716static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
 717{
 718	this_cpu_inc(sch->cpu_qstats->qlen);
 719}
 720
 721static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
 722{
 723	this_cpu_dec(sch->cpu_qstats->qlen);
 724}
 725
 726static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
 727{
 728	this_cpu_inc(sch->cpu_qstats->requeues);
 729}
 730
 731static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
 732{
 733	sch->qstats.drops += count;
 734}
 735
 736static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
 737{
 738	qstats->drops++;
 739}
 740
 741static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
 742{
 743	qstats->overlimits++;
 744}
 745
 746static inline void qdisc_qstats_drop(struct Qdisc *sch)
 747{
 748	qstats_drop_inc(&sch->qstats);
 749}
 750
 751static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
 752{
 753	this_cpu_inc(sch->cpu_qstats->drops);
 754}
 755
 756static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
 757{
 758	sch->qstats.overlimits++;
 759}
 760
 761static inline void qdisc_skb_head_init(struct qdisc_skb_head *qh)
 
 
 
 
 
 
 
 
 762{
 763	qh->head = NULL;
 764	qh->tail = NULL;
 765	qh->qlen = 0;
 
 
 
 
 
 
 
 
 
 
 766}
 767
 768static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
 769				       struct qdisc_skb_head *qh)
 
 
 
 
 
 
 
 
 
 770{
 771	struct sk_buff *last = qh->tail;
 772
 773	if (last) {
 774		skb->next = NULL;
 775		last->next = skb;
 776		qh->tail = skb;
 777	} else {
 778		qh->tail = skb;
 779		qh->head = skb;
 780	}
 781	qh->qlen++;
 782	qdisc_qstats_backlog_inc(sch, skb);
 783
 
 
 
 
 784	return NET_XMIT_SUCCESS;
 785}
 786
 787static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
 
 788{
 789	return __qdisc_enqueue_tail(skb, sch, &sch->q);
 
 
 
 
 
 790}
 791
 792static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
 793{
 794	struct sk_buff *skb = qh->head;
 795
 796	if (likely(skb != NULL)) {
 797		qh->head = skb->next;
 798		qh->qlen--;
 799		if (qh->head == NULL)
 800			qh->tail = NULL;
 801		skb->next = NULL;
 802	}
 803
 804	return skb;
 805}
 806
 807static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
 808{
 809	struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
 810
 811	if (likely(skb != NULL)) {
 812		qdisc_qstats_backlog_dec(sch, skb);
 813		qdisc_bstats_update(sch, skb);
 814	}
 815
 816	return skb;
 817}
 818
 819/* Instead of calling kfree_skb() while root qdisc lock is held,
 820 * queue the skb for future freeing at end of __dev_xmit_skb()
 821 */
 822static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
 823{
 824	skb->next = *to_free;
 825	*to_free = skb;
 826}
 827
 828static inline void __qdisc_drop_all(struct sk_buff *skb,
 829				    struct sk_buff **to_free)
 830{
 831	if (skb->prev)
 832		skb->prev->next = *to_free;
 833	else
 834		skb->next = *to_free;
 835	*to_free = skb;
 836}
 837
 838static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
 839						   struct qdisc_skb_head *qh,
 840						   struct sk_buff **to_free)
 841{
 842	struct sk_buff *skb = __qdisc_dequeue_head(qh);
 843
 844	if (likely(skb != NULL)) {
 845		unsigned int len = qdisc_pkt_len(skb);
 846
 847		qdisc_qstats_backlog_dec(sch, skb);
 848		__qdisc_drop(skb, to_free);
 849		return len;
 850	}
 851
 852	return 0;
 853}
 854
 855static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch,
 856						 struct sk_buff **to_free)
 857{
 858	return __qdisc_queue_drop_head(sch, &sch->q, to_free);
 859}
 860
 861static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
 862{
 863	const struct qdisc_skb_head *qh = &sch->q;
 864
 865	return qh->head;
 866}
 867
 868/* generic pseudo peek method for non-work-conserving qdisc */
 869static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
 870{
 871	struct sk_buff *skb = skb_peek(&sch->gso_skb);
 872
 873	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
 874	if (!skb) {
 875		skb = sch->dequeue(sch);
 876
 877		if (skb) {
 878			__skb_queue_head(&sch->gso_skb, skb);
 879			/* it's still part of the queue */
 880			qdisc_qstats_backlog_inc(sch, skb);
 881			sch->q.qlen++;
 882		}
 883	}
 884
 885	return skb;
 886}
 887
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 888/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
 889static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
 890{
 891	struct sk_buff *skb = skb_peek(&sch->gso_skb);
 892
 893	if (skb) {
 894		skb = __skb_dequeue(&sch->gso_skb);
 895		qdisc_qstats_backlog_dec(sch, skb);
 896		sch->q.qlen--;
 
 
 
 
 
 897	} else {
 898		skb = sch->dequeue(sch);
 899	}
 900
 901	return skb;
 902}
 903
 904static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
 905{
 906	/*
 907	 * We do not know the backlog in bytes of this list, it
 908	 * is up to the caller to correct it
 909	 */
 910	ASSERT_RTNL();
 911	if (qh->qlen) {
 912		rtnl_kfree_skbs(qh->head, qh->tail);
 913
 914		qh->head = NULL;
 915		qh->tail = NULL;
 916		qh->qlen = 0;
 917	}
 918}
 919
 920static inline void qdisc_reset_queue(struct Qdisc *sch)
 921{
 922	__qdisc_reset_queue(&sch->q);
 923	sch->qstats.backlog = 0;
 924}
 925
 926static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
 927					  struct Qdisc **pold)
 928{
 929	struct Qdisc *old;
 930
 931	sch_tree_lock(sch);
 932	old = *pold;
 933	*pold = new;
 934	if (old != NULL) {
 935		unsigned int qlen = old->q.qlen;
 936		unsigned int backlog = old->qstats.backlog;
 937
 938		qdisc_reset(old);
 939		qdisc_tree_reduce_backlog(old, qlen, backlog);
 940	}
 941	sch_tree_unlock(sch);
 942
 943	return old;
 944}
 945
 946static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
 947{
 948	rtnl_kfree_skbs(skb, skb);
 949	qdisc_qstats_drop(sch);
 950}
 951
 952static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
 953				 struct sk_buff **to_free)
 954{
 955	__qdisc_drop(skb, to_free);
 956	qdisc_qstats_cpu_drop(sch);
 957
 958	return NET_XMIT_DROP;
 959}
 960
 961static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
 962			     struct sk_buff **to_free)
 963{
 964	__qdisc_drop(skb, to_free);
 965	qdisc_qstats_drop(sch);
 966
 967	return NET_XMIT_DROP;
 968}
 969
 970static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
 971				 struct sk_buff **to_free)
 972{
 973	__qdisc_drop_all(skb, to_free);
 974	qdisc_qstats_drop(sch);
 975
 976	return NET_XMIT_DROP;
 977}
 978
 979/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
 980   long it will take to send a packet given its size.
 981 */
 982static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
 983{
 984	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
 985	if (slot < 0)
 986		slot = 0;
 987	slot >>= rtab->rate.cell_log;
 988	if (slot > 255)
 989		return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
 990	return rtab->data[slot];
 991}
 992
 993struct psched_ratecfg {
 994	u64	rate_bytes_ps; /* bytes per second */
 995	u32	mult;
 996	u16	overhead;
 
 997	u8	linklayer;
 998	u8	shift;
 999};
1000
1001static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1002				unsigned int len)
1003{
1004	len += r->overhead;
1005
 
 
 
1006	if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1007		return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1008
1009	return ((u64)len * r->mult) >> r->shift;
1010}
1011
1012void psched_ratecfg_precompute(struct psched_ratecfg *r,
1013			       const struct tc_ratespec *conf,
1014			       u64 rate64);
1015
1016static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1017					  const struct psched_ratecfg *r)
1018{
1019	memset(res, 0, sizeof(*res));
1020
1021	/* legacy struct tc_ratespec has a 32bit @rate field
1022	 * Qdisc using 64bit rate should add new attributes
1023	 * in order to maintain compatibility.
1024	 */
1025	res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1026
1027	res->overhead = r->overhead;
 
1028	res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1029}
1030
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1031/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1032 * The fast path only needs to access filter list and to update stats
1033 */
1034struct mini_Qdisc {
1035	struct tcf_proto *filter_list;
1036	struct gnet_stats_basic_cpu __percpu *cpu_bstats;
 
1037	struct gnet_stats_queue	__percpu *cpu_qstats;
1038	struct rcu_head rcu;
1039};
1040
1041static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1042						const struct sk_buff *skb)
1043{
1044	bstats_cpu_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1045}
1046
1047static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1048{
1049	this_cpu_inc(miniq->cpu_qstats->drops);
1050}
1051
1052struct mini_Qdisc_pair {
1053	struct mini_Qdisc miniq1;
1054	struct mini_Qdisc miniq2;
1055	struct mini_Qdisc __rcu **p_miniq;
1056};
1057
1058void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1059			  struct tcf_proto *tp_head);
1060void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1061			  struct mini_Qdisc __rcu **p_miniq);
 
 
 
 
 
 
 
 
 
 
 
 
 
1062
1063#endif
v6.2
   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 <linux/mutex.h>
  16#include <linux/rwsem.h>
  17#include <linux/atomic.h>
  18#include <linux/hashtable.h>
  19#include <net/gen_stats.h>
  20#include <net/rtnetlink.h>
  21#include <net/flow_offload.h>
  22
  23struct Qdisc_ops;
  24struct qdisc_walker;
  25struct tcf_walker;
  26struct module;
  27struct bpf_flow_keys;
  28
  29struct qdisc_rate_table {
  30	struct tc_ratespec rate;
  31	u32		data[256];
  32	struct qdisc_rate_table *next;
  33	int		refcnt;
  34};
  35
  36enum qdisc_state_t {
  37	__QDISC_STATE_SCHED,
  38	__QDISC_STATE_DEACTIVATED,
  39	__QDISC_STATE_MISSED,
  40	__QDISC_STATE_DRAINING,
  41};
  42
  43enum qdisc_state2_t {
  44	/* Only for !TCQ_F_NOLOCK qdisc. Never access it directly.
  45	 * Use qdisc_run_begin/end() or qdisc_is_running() instead.
  46	 */
  47	__QDISC_STATE2_RUNNING,
  48};
  49
  50#define QDISC_STATE_MISSED	BIT(__QDISC_STATE_MISSED)
  51#define QDISC_STATE_DRAINING	BIT(__QDISC_STATE_DRAINING)
  52
  53#define QDISC_STATE_NON_EMPTY	(QDISC_STATE_MISSED | \
  54					QDISC_STATE_DRAINING)
  55
  56struct qdisc_size_table {
  57	struct rcu_head		rcu;
  58	struct list_head	list;
  59	struct tc_sizespec	szopts;
  60	int			refcnt;
  61	u16			data[];
  62};
  63
  64/* similar to sk_buff_head, but skb->prev pointer is undefined. */
  65struct qdisc_skb_head {
  66	struct sk_buff	*head;
  67	struct sk_buff	*tail;
  68	__u32		qlen;
  69	spinlock_t	lock;
  70};
  71
  72struct Qdisc {
  73	int 			(*enqueue)(struct sk_buff *skb,
  74					   struct Qdisc *sch,
  75					   struct sk_buff **to_free);
  76	struct sk_buff *	(*dequeue)(struct Qdisc *sch);
  77	unsigned int		flags;
  78#define TCQ_F_BUILTIN		1
  79#define TCQ_F_INGRESS		2
  80#define TCQ_F_CAN_BYPASS	4
  81#define TCQ_F_MQROOT		8
  82#define TCQ_F_ONETXQUEUE	0x10 /* dequeue_skb() can assume all skbs are for
  83				      * q->dev_queue : It can test
  84				      * netif_xmit_frozen_or_stopped() before
  85				      * dequeueing next packet.
  86				      * Its true for MQ/MQPRIO slaves, or non
  87				      * multiqueue device.
  88				      */
  89#define TCQ_F_WARN_NONWC	(1 << 16)
  90#define TCQ_F_CPUSTATS		0x20 /* run using percpu statistics */
  91#define TCQ_F_NOPARENT		0x40 /* root of its hierarchy :
  92				      * qdisc_tree_decrease_qlen() should stop.
  93				      */
  94#define TCQ_F_INVISIBLE		0x80 /* invisible by default in dump */
  95#define TCQ_F_NOLOCK		0x100 /* qdisc does not require locking */
  96#define TCQ_F_OFFLOADED		0x200 /* qdisc is offloaded to HW */
  97	u32			limit;
  98	const struct Qdisc_ops	*ops;
  99	struct qdisc_size_table	__rcu *stab;
 100	struct hlist_node       hash;
 101	u32			handle;
 102	u32			parent;
 103
 104	struct netdev_queue	*dev_queue;
 105
 106	struct net_rate_estimator __rcu *rate_est;
 107	struct gnet_stats_basic_sync __percpu *cpu_bstats;
 108	struct gnet_stats_queue	__percpu *cpu_qstats;
 109	int			pad;
 110	refcount_t		refcnt;
 111
 112	/*
 113	 * For performance sake on SMP, we put highly modified fields at the end
 114	 */
 115	struct sk_buff_head	gso_skb ____cacheline_aligned_in_smp;
 116	struct qdisc_skb_head	q;
 117	struct gnet_stats_basic_sync bstats;
 
 118	struct gnet_stats_queue	qstats;
 119	unsigned long		state;
 120	unsigned long		state2; /* must be written under qdisc spinlock */
 121	struct Qdisc            *next_sched;
 122	struct sk_buff_head	skb_bad_txq;
 
 
 123
 124	spinlock_t		busylock ____cacheline_aligned_in_smp;
 125	spinlock_t		seqlock;
 126
 127	struct rcu_head		rcu;
 128	netdevice_tracker	dev_tracker;
 129	/* private data */
 130	long privdata[] ____cacheline_aligned;
 131};
 132
 133static inline void qdisc_refcount_inc(struct Qdisc *qdisc)
 134{
 135	if (qdisc->flags & TCQ_F_BUILTIN)
 136		return;
 137	refcount_inc(&qdisc->refcnt);
 138}
 139
 140/* Intended to be used by unlocked users, when concurrent qdisc release is
 141 * possible.
 142 */
 143
 144static inline struct Qdisc *qdisc_refcount_inc_nz(struct Qdisc *qdisc)
 145{
 146	if (qdisc->flags & TCQ_F_BUILTIN)
 147		return qdisc;
 148	if (refcount_inc_not_zero(&qdisc->refcnt))
 149		return qdisc;
 150	return NULL;
 151}
 152
 153/* For !TCQ_F_NOLOCK qdisc: callers must either call this within a qdisc
 154 * root_lock section, or provide their own memory barriers -- ordering
 155 * against qdisc_run_begin/end() atomic bit operations.
 156 */
 157static inline bool qdisc_is_running(struct Qdisc *qdisc)
 158{
 159	if (qdisc->flags & TCQ_F_NOLOCK)
 160		return spin_is_locked(&qdisc->seqlock);
 161	return test_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
 162}
 163
 164static inline bool nolock_qdisc_is_empty(const struct Qdisc *qdisc)
 165{
 166	return !(READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY);
 167}
 168
 169static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
 170{
 171	return q->flags & TCQ_F_CPUSTATS;
 172}
 173
 174static inline bool qdisc_is_empty(const struct Qdisc *qdisc)
 175{
 176	if (qdisc_is_percpu_stats(qdisc))
 177		return nolock_qdisc_is_empty(qdisc);
 178	return !READ_ONCE(qdisc->q.qlen);
 179}
 180
 181/* For !TCQ_F_NOLOCK qdisc, qdisc_run_begin/end() must be invoked with
 182 * the qdisc root lock acquired.
 183 */
 184static inline bool qdisc_run_begin(struct Qdisc *qdisc)
 185{
 186	if (qdisc->flags & TCQ_F_NOLOCK) {
 187		if (spin_trylock(&qdisc->seqlock))
 188			return true;
 189
 190		/* No need to insist if the MISSED flag was already set.
 191		 * Note that test_and_set_bit() also gives us memory ordering
 192		 * guarantees wrt potential earlier enqueue() and below
 193		 * spin_trylock(), both of which are necessary to prevent races
 194		 */
 195		if (test_and_set_bit(__QDISC_STATE_MISSED, &qdisc->state))
 196			return false;
 197
 198		/* Try to take the lock again to make sure that we will either
 199		 * grab it or the CPU that still has it will see MISSED set
 200		 * when testing it in qdisc_run_end()
 201		 */
 202		return spin_trylock(&qdisc->seqlock);
 203	}
 204	return !__test_and_set_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
 205}
 206
 207static inline void qdisc_run_end(struct Qdisc *qdisc)
 208{
 209	if (qdisc->flags & TCQ_F_NOLOCK) {
 210		spin_unlock(&qdisc->seqlock);
 211
 212		/* spin_unlock() only has store-release semantic. The unlock
 213		 * and test_bit() ordering is a store-load ordering, so a full
 214		 * memory barrier is needed here.
 215		 */
 216		smp_mb();
 217
 218		if (unlikely(test_bit(__QDISC_STATE_MISSED,
 219				      &qdisc->state)))
 220			__netif_schedule(qdisc);
 221	} else {
 222		__clear_bit(__QDISC_STATE2_RUNNING, &qdisc->state2);
 223	}
 224}
 225
 226static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
 227{
 228	return qdisc->flags & TCQ_F_ONETXQUEUE;
 229}
 230
 231static inline int qdisc_avail_bulklimit(const struct netdev_queue *txq)
 232{
 233#ifdef CONFIG_BQL
 234	/* Non-BQL migrated drivers will return 0, too. */
 235	return dql_avail(&txq->dql);
 236#else
 237	return 0;
 238#endif
 239}
 240
 241struct Qdisc_class_ops {
 242	unsigned int		flags;
 243	/* Child qdisc manipulation */
 244	struct netdev_queue *	(*select_queue)(struct Qdisc *, struct tcmsg *);
 245	int			(*graft)(struct Qdisc *, unsigned long cl,
 246					struct Qdisc *, struct Qdisc **,
 247					struct netlink_ext_ack *extack);
 248	struct Qdisc *		(*leaf)(struct Qdisc *, unsigned long cl);
 249	void			(*qlen_notify)(struct Qdisc *, unsigned long);
 250
 251	/* Class manipulation routines */
 252	unsigned long		(*find)(struct Qdisc *, u32 classid);
 253	int			(*change)(struct Qdisc *, u32, u32,
 254					struct nlattr **, unsigned long *,
 255					struct netlink_ext_ack *);
 256	int			(*delete)(struct Qdisc *, unsigned long,
 257					  struct netlink_ext_ack *);
 258	void			(*walk)(struct Qdisc *, struct qdisc_walker * arg);
 259
 260	/* Filter manipulation */
 261	struct tcf_block *	(*tcf_block)(struct Qdisc *sch,
 262					     unsigned long arg,
 263					     struct netlink_ext_ack *extack);
 264	unsigned long		(*bind_tcf)(struct Qdisc *, unsigned long,
 265					u32 classid);
 266	void			(*unbind_tcf)(struct Qdisc *, unsigned long);
 267
 268	/* rtnetlink specific */
 269	int			(*dump)(struct Qdisc *, unsigned long,
 270					struct sk_buff *skb, struct tcmsg*);
 271	int			(*dump_stats)(struct Qdisc *, unsigned long,
 272					struct gnet_dump *);
 273};
 274
 275/* Qdisc_class_ops flag values */
 276
 277/* Implements API that doesn't require rtnl lock */
 278enum qdisc_class_ops_flags {
 279	QDISC_CLASS_OPS_DOIT_UNLOCKED = 1,
 280};
 281
 282struct Qdisc_ops {
 283	struct Qdisc_ops	*next;
 284	const struct Qdisc_class_ops	*cl_ops;
 285	char			id[IFNAMSIZ];
 286	int			priv_size;
 287	unsigned int		static_flags;
 288
 289	int 			(*enqueue)(struct sk_buff *skb,
 290					   struct Qdisc *sch,
 291					   struct sk_buff **to_free);
 292	struct sk_buff *	(*dequeue)(struct Qdisc *);
 293	struct sk_buff *	(*peek)(struct Qdisc *);
 294
 295	int			(*init)(struct Qdisc *sch, struct nlattr *arg,
 296					struct netlink_ext_ack *extack);
 297	void			(*reset)(struct Qdisc *);
 298	void			(*destroy)(struct Qdisc *);
 299	int			(*change)(struct Qdisc *sch,
 300					  struct nlattr *arg,
 301					  struct netlink_ext_ack *extack);
 302	void			(*attach)(struct Qdisc *sch);
 303	int			(*change_tx_queue_len)(struct Qdisc *, unsigned int);
 304	void			(*change_real_num_tx)(struct Qdisc *sch,
 305						      unsigned int new_real_tx);
 306
 307	int			(*dump)(struct Qdisc *, struct sk_buff *);
 308	int			(*dump_stats)(struct Qdisc *, struct gnet_dump *);
 309
 310	void			(*ingress_block_set)(struct Qdisc *sch,
 311						     u32 block_index);
 312	void			(*egress_block_set)(struct Qdisc *sch,
 313						    u32 block_index);
 314	u32			(*ingress_block_get)(struct Qdisc *sch);
 315	u32			(*egress_block_get)(struct Qdisc *sch);
 316
 317	struct module		*owner;
 318};
 319
 320
 321struct tcf_result {
 322	union {
 323		struct {
 324			unsigned long	class;
 325			u32		classid;
 326		};
 327		const struct tcf_proto *goto_tp;
 328
 329	};
 330};
 331
 332struct tcf_chain;
 333
 334struct tcf_proto_ops {
 335	struct list_head	head;
 336	char			kind[IFNAMSIZ];
 337
 338	int			(*classify)(struct sk_buff *,
 339					    const struct tcf_proto *,
 340					    struct tcf_result *);
 341	int			(*init)(struct tcf_proto*);
 342	void			(*destroy)(struct tcf_proto *tp, bool rtnl_held,
 343					   struct netlink_ext_ack *extack);
 344
 345	void*			(*get)(struct tcf_proto*, u32 handle);
 346	void			(*put)(struct tcf_proto *tp, void *f);
 347	int			(*change)(struct net *net, struct sk_buff *,
 348					struct tcf_proto*, unsigned long,
 349					u32 handle, struct nlattr **,
 350					void **, u32,
 351					struct netlink_ext_ack *);
 352	int			(*delete)(struct tcf_proto *tp, void *arg,
 353					  bool *last, bool rtnl_held,
 354					  struct netlink_ext_ack *);
 355	bool			(*delete_empty)(struct tcf_proto *tp);
 356	void			(*walk)(struct tcf_proto *tp,
 357					struct tcf_walker *arg, bool rtnl_held);
 358	int			(*reoffload)(struct tcf_proto *tp, bool add,
 359					     flow_setup_cb_t *cb, void *cb_priv,
 360					     struct netlink_ext_ack *extack);
 361	void			(*hw_add)(struct tcf_proto *tp,
 362					  void *type_data);
 363	void			(*hw_del)(struct tcf_proto *tp,
 364					  void *type_data);
 365	void			(*bind_class)(void *, u32, unsigned long,
 366					      void *, unsigned long);
 367	void *			(*tmplt_create)(struct net *net,
 368						struct tcf_chain *chain,
 369						struct nlattr **tca,
 370						struct netlink_ext_ack *extack);
 371	void			(*tmplt_destroy)(void *tmplt_priv);
 372
 373	/* rtnetlink specific */
 374	int			(*dump)(struct net*, struct tcf_proto*, void *,
 375					struct sk_buff *skb, struct tcmsg*,
 376					bool);
 377	int			(*terse_dump)(struct net *net,
 378					      struct tcf_proto *tp, void *fh,
 379					      struct sk_buff *skb,
 380					      struct tcmsg *t, bool rtnl_held);
 381	int			(*tmplt_dump)(struct sk_buff *skb,
 382					      struct net *net,
 383					      void *tmplt_priv);
 384
 385	struct module		*owner;
 386	int			flags;
 387};
 388
 389/* Classifiers setting TCF_PROTO_OPS_DOIT_UNLOCKED in tcf_proto_ops->flags
 390 * are expected to implement tcf_proto_ops->delete_empty(), otherwise race
 391 * conditions can occur when filters are inserted/deleted simultaneously.
 392 */
 393enum tcf_proto_ops_flags {
 394	TCF_PROTO_OPS_DOIT_UNLOCKED = 1,
 395};
 396
 397struct tcf_proto {
 398	/* Fast access part */
 399	struct tcf_proto __rcu	*next;
 400	void __rcu		*root;
 401
 402	/* called under RCU BH lock*/
 403	int			(*classify)(struct sk_buff *,
 404					    const struct tcf_proto *,
 405					    struct tcf_result *);
 406	__be16			protocol;
 407
 408	/* All the rest */
 409	u32			prio;
 410	void			*data;
 411	const struct tcf_proto_ops	*ops;
 412	struct tcf_chain	*chain;
 413	/* Lock protects tcf_proto shared state and can be used by unlocked
 414	 * classifiers to protect their private data.
 415	 */
 416	spinlock_t		lock;
 417	bool			deleting;
 418	refcount_t		refcnt;
 419	struct rcu_head		rcu;
 420	struct hlist_node	destroy_ht_node;
 421};
 422
 423struct qdisc_skb_cb {
 424	struct {
 425		unsigned int		pkt_len;
 426		u16			slave_dev_queue_mapping;
 427		u16			tc_classid;
 428	};
 429#define QDISC_CB_PRIV_LEN 20
 430	unsigned char		data[QDISC_CB_PRIV_LEN];
 431};
 432
 433typedef void tcf_chain_head_change_t(struct tcf_proto *tp_head, void *priv);
 434
 435struct tcf_chain {
 436	/* Protects filter_chain. */
 437	struct mutex filter_chain_lock;
 438	struct tcf_proto __rcu *filter_chain;
 
 439	struct list_head list;
 440	struct tcf_block *block;
 441	u32 index; /* chain index */
 442	unsigned int refcnt;
 443	unsigned int action_refcnt;
 444	bool explicitly_created;
 445	bool flushing;
 446	const struct tcf_proto_ops *tmplt_ops;
 447	void *tmplt_priv;
 448	struct rcu_head rcu;
 449};
 450
 451struct tcf_block {
 452	/* Lock protects tcf_block and lifetime-management data of chains
 453	 * attached to the block (refcnt, action_refcnt, explicitly_created).
 454	 */
 455	struct mutex lock;
 456	struct list_head chain_list;
 457	u32 index; /* block index for shared blocks */
 458	u32 classid; /* which class this block belongs to */
 459	refcount_t refcnt;
 460	struct net *net;
 461	struct Qdisc *q;
 462	struct rw_semaphore cb_lock; /* protects cb_list and offload counters */
 463	struct flow_block flow_block;
 464	struct list_head owner_list;
 465	bool keep_dst;
 466	atomic_t offloadcnt; /* Number of oddloaded filters */
 467	unsigned int nooffloaddevcnt; /* Number of devs unable to do offload */
 468	unsigned int lockeddevcnt; /* Number of devs that require rtnl lock. */
 469	struct {
 470		struct tcf_chain *chain;
 471		struct list_head filter_chain_list;
 472	} chain0;
 473	struct rcu_head rcu;
 474	DECLARE_HASHTABLE(proto_destroy_ht, 7);
 475	struct mutex proto_destroy_lock; /* Lock for proto_destroy hashtable. */
 476};
 477
 478static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain *chain)
 479{
 480	return lockdep_is_held(&chain->filter_chain_lock);
 
 
 
 481}
 482
 483static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto *tp)
 484{
 485	return lockdep_is_held(&tp->lock);
 
 
 
 486}
 487
 488#define tcf_chain_dereference(p, chain)					\
 489	rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
 490
 491#define tcf_proto_dereference(p, tp)					\
 492	rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
 493
 494static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz)
 495{
 496	struct qdisc_skb_cb *qcb;
 497
 498	BUILD_BUG_ON(sizeof(skb->cb) < sizeof(*qcb));
 499	BUILD_BUG_ON(sizeof(qcb->data) < sz);
 500}
 501
 
 
 
 
 
 502static inline int qdisc_qlen(const struct Qdisc *q)
 503{
 504	return q->q.qlen;
 505}
 506
 507static inline int qdisc_qlen_sum(const struct Qdisc *q)
 508{
 509	__u32 qlen = q->qstats.qlen;
 510	int i;
 511
 512	if (qdisc_is_percpu_stats(q)) {
 513		for_each_possible_cpu(i)
 514			qlen += per_cpu_ptr(q->cpu_qstats, i)->qlen;
 515	} else {
 516		qlen += q->q.qlen;
 517	}
 518
 519	return qlen;
 520}
 521
 522static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb)
 523{
 524	return (struct qdisc_skb_cb *)skb->cb;
 525}
 526
 527static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
 528{
 529	return &qdisc->q.lock;
 530}
 531
 532static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc)
 533{
 534	struct Qdisc *q = rcu_dereference_rtnl(qdisc->dev_queue->qdisc);
 535
 536	return q;
 537}
 538
 539static inline struct Qdisc *qdisc_root_bh(const struct Qdisc *qdisc)
 540{
 541	return rcu_dereference_bh(qdisc->dev_queue->qdisc);
 542}
 543
 544static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc)
 
 
 
 
 
 
 
 
 
 
 
 545{
 546	return qdisc->dev_queue->qdisc_sleeping;
 
 
 
 547}
 548
 549static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
 550{
 551	struct Qdisc *root = qdisc_root_sleeping(qdisc);
 552
 553	ASSERT_RTNL();
 554	return qdisc_lock(root);
 555}
 556
 
 
 
 
 
 
 
 
 557static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
 558{
 559	return qdisc->dev_queue->dev;
 560}
 561
 562static inline void sch_tree_lock(struct Qdisc *q)
 563{
 564	if (q->flags & TCQ_F_MQROOT)
 565		spin_lock_bh(qdisc_lock(q));
 566	else
 567		spin_lock_bh(qdisc_root_sleeping_lock(q));
 568}
 569
 570static inline void sch_tree_unlock(struct Qdisc *q)
 571{
 572	if (q->flags & TCQ_F_MQROOT)
 573		spin_unlock_bh(qdisc_lock(q));
 574	else
 575		spin_unlock_bh(qdisc_root_sleeping_lock(q));
 576}
 577
 578extern struct Qdisc noop_qdisc;
 579extern struct Qdisc_ops noop_qdisc_ops;
 580extern struct Qdisc_ops pfifo_fast_ops;
 581extern struct Qdisc_ops mq_qdisc_ops;
 582extern struct Qdisc_ops noqueue_qdisc_ops;
 583extern const struct Qdisc_ops *default_qdisc_ops;
 584static inline const struct Qdisc_ops *
 585get_default_qdisc_ops(const struct net_device *dev, int ntx)
 586{
 587	return ntx < dev->real_num_tx_queues ?
 588			default_qdisc_ops : &pfifo_fast_ops;
 589}
 590
 591struct Qdisc_class_common {
 592	u32			classid;
 593	struct hlist_node	hnode;
 594};
 595
 596struct Qdisc_class_hash {
 597	struct hlist_head	*hash;
 598	unsigned int		hashsize;
 599	unsigned int		hashmask;
 600	unsigned int		hashelems;
 601};
 602
 603static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
 604{
 605	id ^= id >> 8;
 606	id ^= id >> 4;
 607	return id & mask;
 608}
 609
 610static inline struct Qdisc_class_common *
 611qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id)
 612{
 613	struct Qdisc_class_common *cl;
 614	unsigned int h;
 615
 616	if (!id)
 617		return NULL;
 618
 619	h = qdisc_class_hash(id, hash->hashmask);
 620	hlist_for_each_entry(cl, &hash->hash[h], hnode) {
 621		if (cl->classid == id)
 622			return cl;
 623	}
 624	return NULL;
 625}
 626
 627static inline int tc_classid_to_hwtc(struct net_device *dev, u32 classid)
 628{
 629	u32 hwtc = TC_H_MIN(classid) - TC_H_MIN_PRIORITY;
 630
 631	return (hwtc < netdev_get_num_tc(dev)) ? hwtc : -EINVAL;
 632}
 633
 634int qdisc_class_hash_init(struct Qdisc_class_hash *);
 635void qdisc_class_hash_insert(struct Qdisc_class_hash *,
 636			     struct Qdisc_class_common *);
 637void qdisc_class_hash_remove(struct Qdisc_class_hash *,
 638			     struct Qdisc_class_common *);
 639void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
 640void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
 641
 642int dev_qdisc_change_tx_queue_len(struct net_device *dev);
 643void dev_qdisc_change_real_num_tx(struct net_device *dev,
 644				  unsigned int new_real_tx);
 645void dev_init_scheduler(struct net_device *dev);
 646void dev_shutdown(struct net_device *dev);
 647void dev_activate(struct net_device *dev);
 648void dev_deactivate(struct net_device *dev);
 649void dev_deactivate_many(struct list_head *head);
 650struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
 651			      struct Qdisc *qdisc);
 652void qdisc_reset(struct Qdisc *qdisc);
 653void qdisc_put(struct Qdisc *qdisc);
 654void qdisc_put_unlocked(struct Qdisc *qdisc);
 655void qdisc_tree_reduce_backlog(struct Qdisc *qdisc, int n, int len);
 656#ifdef CONFIG_NET_SCHED
 657int qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
 658			      void *type_data);
 659void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
 660				struct Qdisc *new, struct Qdisc *old,
 661				enum tc_setup_type type, void *type_data,
 662				struct netlink_ext_ack *extack);
 663#else
 664static inline int
 665qdisc_offload_dump_helper(struct Qdisc *q, enum tc_setup_type type,
 666			  void *type_data)
 667{
 668	q->flags &= ~TCQ_F_OFFLOADED;
 669	return 0;
 670}
 671
 672static inline void
 673qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
 674			   struct Qdisc *new, struct Qdisc *old,
 675			   enum tc_setup_type type, void *type_data,
 676			   struct netlink_ext_ack *extack)
 677{
 678}
 679#endif
 680void qdisc_offload_query_caps(struct net_device *dev,
 681			      enum tc_setup_type type,
 682			      void *caps, size_t caps_len);
 683struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
 684			  const struct Qdisc_ops *ops,
 685			  struct netlink_ext_ack *extack);
 686void qdisc_free(struct Qdisc *qdisc);
 687struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
 688				const struct Qdisc_ops *ops, u32 parentid,
 689				struct netlink_ext_ack *extack);
 690void __qdisc_calculate_pkt_len(struct sk_buff *skb,
 691			       const struct qdisc_size_table *stab);
 692int skb_do_redirect(struct sk_buff *);
 693
 
 
 
 
 
 
 
 694static inline bool skb_at_tc_ingress(const struct sk_buff *skb)
 695{
 696#ifdef CONFIG_NET_CLS_ACT
 697	return skb->tc_at_ingress;
 698#else
 699	return false;
 700#endif
 701}
 702
 703static inline bool skb_skip_tc_classify(struct sk_buff *skb)
 704{
 705#ifdef CONFIG_NET_CLS_ACT
 706	if (skb->tc_skip_classify) {
 707		skb->tc_skip_classify = 0;
 708		return true;
 709	}
 710#endif
 711	return false;
 712}
 713
 714/* Reset all TX qdiscs greater than index of a device.  */
 715static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i)
 716{
 717	struct Qdisc *qdisc;
 718
 719	for (; i < dev->num_tx_queues; i++) {
 720		qdisc = rtnl_dereference(netdev_get_tx_queue(dev, i)->qdisc);
 721		if (qdisc) {
 722			spin_lock_bh(qdisc_lock(qdisc));
 723			qdisc_reset(qdisc);
 724			spin_unlock_bh(qdisc_lock(qdisc));
 725		}
 726	}
 727}
 728
 
 
 
 
 
 729/* Are all TX queues of the device empty?  */
 730static inline bool qdisc_all_tx_empty(const struct net_device *dev)
 731{
 732	unsigned int i;
 733
 734	rcu_read_lock();
 735	for (i = 0; i < dev->num_tx_queues; i++) {
 736		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 737		const struct Qdisc *q = rcu_dereference(txq->qdisc);
 738
 739		if (!qdisc_is_empty(q)) {
 740			rcu_read_unlock();
 741			return false;
 742		}
 743	}
 744	rcu_read_unlock();
 745	return true;
 746}
 747
 748/* Are any of the TX qdiscs changing?  */
 749static inline bool qdisc_tx_changing(const struct net_device *dev)
 750{
 751	unsigned int i;
 752
 753	for (i = 0; i < dev->num_tx_queues; i++) {
 754		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 755		if (rcu_access_pointer(txq->qdisc) != txq->qdisc_sleeping)
 756			return true;
 757	}
 758	return false;
 759}
 760
 761/* Is the device using the noop qdisc on all queues?  */
 762static inline bool qdisc_tx_is_noop(const struct net_device *dev)
 763{
 764	unsigned int i;
 765
 766	for (i = 0; i < dev->num_tx_queues; i++) {
 767		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 768		if (rcu_access_pointer(txq->qdisc) != &noop_qdisc)
 769			return false;
 770	}
 771	return true;
 772}
 773
 774static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb)
 775{
 776	return qdisc_skb_cb(skb)->pkt_len;
 777}
 778
 779/* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
 780enum net_xmit_qdisc_t {
 781	__NET_XMIT_STOLEN = 0x00010000,
 782	__NET_XMIT_BYPASS = 0x00020000,
 783};
 784
 785#ifdef CONFIG_NET_CLS_ACT
 786#define net_xmit_drop_count(e)	((e) & __NET_XMIT_STOLEN ? 0 : 1)
 787#else
 788#define net_xmit_drop_count(e)	(1)
 789#endif
 790
 791static inline void qdisc_calculate_pkt_len(struct sk_buff *skb,
 792					   const struct Qdisc *sch)
 793{
 794#ifdef CONFIG_NET_SCHED
 795	struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab);
 796
 797	if (stab)
 798		__qdisc_calculate_pkt_len(skb, stab);
 799#endif
 800}
 801
 802static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 803				struct sk_buff **to_free)
 804{
 805	qdisc_calculate_pkt_len(skb, sch);
 806	return sch->enqueue(skb, sch, to_free);
 807}
 808
 809static inline void _bstats_update(struct gnet_stats_basic_sync *bstats,
 
 
 
 
 
 810				  __u64 bytes, __u32 packets)
 811{
 812	u64_stats_update_begin(&bstats->syncp);
 813	u64_stats_add(&bstats->bytes, bytes);
 814	u64_stats_add(&bstats->packets, packets);
 815	u64_stats_update_end(&bstats->syncp);
 816}
 817
 818static inline void bstats_update(struct gnet_stats_basic_sync *bstats,
 819				 const struct sk_buff *skb)
 820{
 821	_bstats_update(bstats,
 822		       qdisc_pkt_len(skb),
 823		       skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1);
 824}
 825
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 826static inline void qdisc_bstats_cpu_update(struct Qdisc *sch,
 827					   const struct sk_buff *skb)
 828{
 829	bstats_update(this_cpu_ptr(sch->cpu_bstats), skb);
 830}
 831
 832static inline void qdisc_bstats_update(struct Qdisc *sch,
 833				       const struct sk_buff *skb)
 834{
 835	bstats_update(&sch->bstats, skb);
 836}
 837
 838static inline void qdisc_qstats_backlog_dec(struct Qdisc *sch,
 839					    const struct sk_buff *skb)
 840{
 841	sch->qstats.backlog -= qdisc_pkt_len(skb);
 842}
 843
 844static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc *sch,
 845						const struct sk_buff *skb)
 846{
 847	this_cpu_sub(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
 848}
 849
 850static inline void qdisc_qstats_backlog_inc(struct Qdisc *sch,
 851					    const struct sk_buff *skb)
 852{
 853	sch->qstats.backlog += qdisc_pkt_len(skb);
 854}
 855
 856static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc *sch,
 857						const struct sk_buff *skb)
 858{
 859	this_cpu_add(sch->cpu_qstats->backlog, qdisc_pkt_len(skb));
 860}
 861
 862static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc *sch)
 863{
 864	this_cpu_inc(sch->cpu_qstats->qlen);
 865}
 866
 867static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc *sch)
 868{
 869	this_cpu_dec(sch->cpu_qstats->qlen);
 870}
 871
 872static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc *sch)
 873{
 874	this_cpu_inc(sch->cpu_qstats->requeues);
 875}
 876
 877static inline void __qdisc_qstats_drop(struct Qdisc *sch, int count)
 878{
 879	sch->qstats.drops += count;
 880}
 881
 882static inline void qstats_drop_inc(struct gnet_stats_queue *qstats)
 883{
 884	qstats->drops++;
 885}
 886
 887static inline void qstats_overlimit_inc(struct gnet_stats_queue *qstats)
 888{
 889	qstats->overlimits++;
 890}
 891
 892static inline void qdisc_qstats_drop(struct Qdisc *sch)
 893{
 894	qstats_drop_inc(&sch->qstats);
 895}
 896
 897static inline void qdisc_qstats_cpu_drop(struct Qdisc *sch)
 898{
 899	this_cpu_inc(sch->cpu_qstats->drops);
 900}
 901
 902static inline void qdisc_qstats_overlimit(struct Qdisc *sch)
 903{
 904	sch->qstats.overlimits++;
 905}
 906
 907static inline int qdisc_qstats_copy(struct gnet_dump *d, struct Qdisc *sch)
 908{
 909	__u32 qlen = qdisc_qlen_sum(sch);
 910
 911	return gnet_stats_copy_queue(d, sch->cpu_qstats, &sch->qstats, qlen);
 912}
 913
 914static inline void qdisc_qstats_qlen_backlog(struct Qdisc *sch,  __u32 *qlen,
 915					     __u32 *backlog)
 916{
 917	struct gnet_stats_queue qstats = { 0 };
 918
 919	gnet_stats_add_queue(&qstats, sch->cpu_qstats, &sch->qstats);
 920	*qlen = qstats.qlen + qdisc_qlen(sch);
 921	*backlog = qstats.backlog;
 922}
 923
 924static inline void qdisc_tree_flush_backlog(struct Qdisc *sch)
 925{
 926	__u32 qlen, backlog;
 927
 928	qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
 929	qdisc_tree_reduce_backlog(sch, qlen, backlog);
 930}
 931
 932static inline void qdisc_purge_queue(struct Qdisc *sch)
 933{
 934	__u32 qlen, backlog;
 935
 936	qdisc_qstats_qlen_backlog(sch, &qlen, &backlog);
 937	qdisc_reset(sch);
 938	qdisc_tree_reduce_backlog(sch, qlen, backlog);
 939}
 940
 941static inline void __qdisc_enqueue_tail(struct sk_buff *skb,
 942					struct qdisc_skb_head *qh)
 943{
 944	struct sk_buff *last = qh->tail;
 945
 946	if (last) {
 947		skb->next = NULL;
 948		last->next = skb;
 949		qh->tail = skb;
 950	} else {
 951		qh->tail = skb;
 952		qh->head = skb;
 953	}
 954	qh->qlen++;
 955}
 956
 957static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
 958{
 959	__qdisc_enqueue_tail(skb, &sch->q);
 960	qdisc_qstats_backlog_inc(sch, skb);
 961	return NET_XMIT_SUCCESS;
 962}
 963
 964static inline void __qdisc_enqueue_head(struct sk_buff *skb,
 965					struct qdisc_skb_head *qh)
 966{
 967	skb->next = qh->head;
 968
 969	if (!qh->head)
 970		qh->tail = skb;
 971	qh->head = skb;
 972	qh->qlen++;
 973}
 974
 975static inline struct sk_buff *__qdisc_dequeue_head(struct qdisc_skb_head *qh)
 976{
 977	struct sk_buff *skb = qh->head;
 978
 979	if (likely(skb != NULL)) {
 980		qh->head = skb->next;
 981		qh->qlen--;
 982		if (qh->head == NULL)
 983			qh->tail = NULL;
 984		skb->next = NULL;
 985	}
 986
 987	return skb;
 988}
 989
 990static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
 991{
 992	struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
 993
 994	if (likely(skb != NULL)) {
 995		qdisc_qstats_backlog_dec(sch, skb);
 996		qdisc_bstats_update(sch, skb);
 997	}
 998
 999	return skb;
1000}
1001
1002/* Instead of calling kfree_skb() while root qdisc lock is held,
1003 * queue the skb for future freeing at end of __dev_xmit_skb()
1004 */
1005static inline void __qdisc_drop(struct sk_buff *skb, struct sk_buff **to_free)
1006{
1007	skb->next = *to_free;
1008	*to_free = skb;
1009}
1010
1011static inline void __qdisc_drop_all(struct sk_buff *skb,
1012				    struct sk_buff **to_free)
1013{
1014	if (skb->prev)
1015		skb->prev->next = *to_free;
1016	else
1017		skb->next = *to_free;
1018	*to_free = skb;
1019}
1020
1021static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch,
1022						   struct qdisc_skb_head *qh,
1023						   struct sk_buff **to_free)
1024{
1025	struct sk_buff *skb = __qdisc_dequeue_head(qh);
1026
1027	if (likely(skb != NULL)) {
1028		unsigned int len = qdisc_pkt_len(skb);
1029
1030		qdisc_qstats_backlog_dec(sch, skb);
1031		__qdisc_drop(skb, to_free);
1032		return len;
1033	}
1034
1035	return 0;
1036}
1037
 
 
 
 
 
 
1038static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
1039{
1040	const struct qdisc_skb_head *qh = &sch->q;
1041
1042	return qh->head;
1043}
1044
1045/* generic pseudo peek method for non-work-conserving qdisc */
1046static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
1047{
1048	struct sk_buff *skb = skb_peek(&sch->gso_skb);
1049
1050	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1051	if (!skb) {
1052		skb = sch->dequeue(sch);
1053
1054		if (skb) {
1055			__skb_queue_head(&sch->gso_skb, skb);
1056			/* it's still part of the queue */
1057			qdisc_qstats_backlog_inc(sch, skb);
1058			sch->q.qlen++;
1059		}
1060	}
1061
1062	return skb;
1063}
1064
1065static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
1066						 struct sk_buff *skb)
1067{
1068	if (qdisc_is_percpu_stats(sch)) {
1069		qdisc_qstats_cpu_backlog_dec(sch, skb);
1070		qdisc_bstats_cpu_update(sch, skb);
1071		qdisc_qstats_cpu_qlen_dec(sch);
1072	} else {
1073		qdisc_qstats_backlog_dec(sch, skb);
1074		qdisc_bstats_update(sch, skb);
1075		sch->q.qlen--;
1076	}
1077}
1078
1079static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
1080						 unsigned int pkt_len)
1081{
1082	if (qdisc_is_percpu_stats(sch)) {
1083		qdisc_qstats_cpu_qlen_inc(sch);
1084		this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
1085	} else {
1086		sch->qstats.backlog += pkt_len;
1087		sch->q.qlen++;
1088	}
1089}
1090
1091/* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1092static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
1093{
1094	struct sk_buff *skb = skb_peek(&sch->gso_skb);
1095
1096	if (skb) {
1097		skb = __skb_dequeue(&sch->gso_skb);
1098		if (qdisc_is_percpu_stats(sch)) {
1099			qdisc_qstats_cpu_backlog_dec(sch, skb);
1100			qdisc_qstats_cpu_qlen_dec(sch);
1101		} else {
1102			qdisc_qstats_backlog_dec(sch, skb);
1103			sch->q.qlen--;
1104		}
1105	} else {
1106		skb = sch->dequeue(sch);
1107	}
1108
1109	return skb;
1110}
1111
1112static inline void __qdisc_reset_queue(struct qdisc_skb_head *qh)
1113{
1114	/*
1115	 * We do not know the backlog in bytes of this list, it
1116	 * is up to the caller to correct it
1117	 */
1118	ASSERT_RTNL();
1119	if (qh->qlen) {
1120		rtnl_kfree_skbs(qh->head, qh->tail);
1121
1122		qh->head = NULL;
1123		qh->tail = NULL;
1124		qh->qlen = 0;
1125	}
1126}
1127
1128static inline void qdisc_reset_queue(struct Qdisc *sch)
1129{
1130	__qdisc_reset_queue(&sch->q);
 
1131}
1132
1133static inline struct Qdisc *qdisc_replace(struct Qdisc *sch, struct Qdisc *new,
1134					  struct Qdisc **pold)
1135{
1136	struct Qdisc *old;
1137
1138	sch_tree_lock(sch);
1139	old = *pold;
1140	*pold = new;
1141	if (old != NULL)
1142		qdisc_purge_queue(old);
 
 
 
 
 
1143	sch_tree_unlock(sch);
1144
1145	return old;
1146}
1147
1148static inline void rtnl_qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
1149{
1150	rtnl_kfree_skbs(skb, skb);
1151	qdisc_qstats_drop(sch);
1152}
1153
1154static inline int qdisc_drop_cpu(struct sk_buff *skb, struct Qdisc *sch,
1155				 struct sk_buff **to_free)
1156{
1157	__qdisc_drop(skb, to_free);
1158	qdisc_qstats_cpu_drop(sch);
1159
1160	return NET_XMIT_DROP;
1161}
1162
1163static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch,
1164			     struct sk_buff **to_free)
1165{
1166	__qdisc_drop(skb, to_free);
1167	qdisc_qstats_drop(sch);
1168
1169	return NET_XMIT_DROP;
1170}
1171
1172static inline int qdisc_drop_all(struct sk_buff *skb, struct Qdisc *sch,
1173				 struct sk_buff **to_free)
1174{
1175	__qdisc_drop_all(skb, to_free);
1176	qdisc_qstats_drop(sch);
1177
1178	return NET_XMIT_DROP;
1179}
1180
1181/* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1182   long it will take to send a packet given its size.
1183 */
1184static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
1185{
1186	int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
1187	if (slot < 0)
1188		slot = 0;
1189	slot >>= rtab->rate.cell_log;
1190	if (slot > 255)
1191		return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF];
1192	return rtab->data[slot];
1193}
1194
1195struct psched_ratecfg {
1196	u64	rate_bytes_ps; /* bytes per second */
1197	u32	mult;
1198	u16	overhead;
1199	u16	mpu;
1200	u8	linklayer;
1201	u8	shift;
1202};
1203
1204static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
1205				unsigned int len)
1206{
1207	len += r->overhead;
1208
1209	if (len < r->mpu)
1210		len = r->mpu;
1211
1212	if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
1213		return ((u64)(DIV_ROUND_UP(len,48)*53) * r->mult) >> r->shift;
1214
1215	return ((u64)len * r->mult) >> r->shift;
1216}
1217
1218void psched_ratecfg_precompute(struct psched_ratecfg *r,
1219			       const struct tc_ratespec *conf,
1220			       u64 rate64);
1221
1222static inline void psched_ratecfg_getrate(struct tc_ratespec *res,
1223					  const struct psched_ratecfg *r)
1224{
1225	memset(res, 0, sizeof(*res));
1226
1227	/* legacy struct tc_ratespec has a 32bit @rate field
1228	 * Qdisc using 64bit rate should add new attributes
1229	 * in order to maintain compatibility.
1230	 */
1231	res->rate = min_t(u64, r->rate_bytes_ps, ~0U);
1232
1233	res->overhead = r->overhead;
1234	res->mpu = r->mpu;
1235	res->linklayer = (r->linklayer & TC_LINKLAYER_MASK);
1236}
1237
1238struct psched_pktrate {
1239	u64	rate_pkts_ps; /* packets per second */
1240	u32	mult;
1241	u8	shift;
1242};
1243
1244static inline u64 psched_pkt2t_ns(const struct psched_pktrate *r,
1245				  unsigned int pkt_num)
1246{
1247	return ((u64)pkt_num * r->mult) >> r->shift;
1248}
1249
1250void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64);
1251
1252/* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1253 * The fast path only needs to access filter list and to update stats
1254 */
1255struct mini_Qdisc {
1256	struct tcf_proto *filter_list;
1257	struct tcf_block *block;
1258	struct gnet_stats_basic_sync __percpu *cpu_bstats;
1259	struct gnet_stats_queue	__percpu *cpu_qstats;
1260	unsigned long rcu_state;
1261};
1262
1263static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc *miniq,
1264						const struct sk_buff *skb)
1265{
1266	bstats_update(this_cpu_ptr(miniq->cpu_bstats), skb);
1267}
1268
1269static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc *miniq)
1270{
1271	this_cpu_inc(miniq->cpu_qstats->drops);
1272}
1273
1274struct mini_Qdisc_pair {
1275	struct mini_Qdisc miniq1;
1276	struct mini_Qdisc miniq2;
1277	struct mini_Qdisc __rcu **p_miniq;
1278};
1279
1280void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1281			  struct tcf_proto *tp_head);
1282void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1283			  struct mini_Qdisc __rcu **p_miniq);
1284void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1285				struct tcf_block *block);
1286
1287void mq_change_real_num_tx(struct Qdisc *sch, unsigned int new_real_tx);
1288
1289int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1290
1291/* Make sure qdisc is no longer in SCHED state. */
1292static inline void qdisc_synchronize(const struct Qdisc *q)
1293{
1294	while (test_bit(__QDISC_STATE_SCHED, &q->state))
1295		msleep(1);
1296}
1297
1298#endif