1 #ifndef __NET_SCHED_GENERIC_H
2 #define __NET_SCHED_GENERIC_H
4 #include <linux/netdevice.h>
5 #include <linux/types.h>
6 #include <linux/rcupdate.h>
7 #include <linux/pkt_sched.h>
8 #include <linux/pkt_cls.h>
9 #include <net/gen_stats.h>
10 #include <net/rtnetlink.h>
17 struct qdisc_rate_table
{
18 struct tc_ratespec rate
;
20 struct qdisc_rate_table
*next
;
26 __QDISC_STATE_DEACTIVATED
,
27 __QDISC_STATE_THROTTLED
,
31 * following bits are only changed while qdisc lock is held
33 enum qdisc___state_t
{
34 __QDISC___STATE_RUNNING
= 1,
37 struct qdisc_size_table
{
39 struct list_head list
;
40 struct tc_sizespec szopts
;
46 int (*enqueue
)(struct sk_buff
*skb
, struct Qdisc
*dev
);
47 struct sk_buff
* (*dequeue
)(struct Qdisc
*dev
);
49 #define TCQ_F_BUILTIN 1
50 #define TCQ_F_INGRESS 2
51 #define TCQ_F_CAN_BYPASS 4
52 #define TCQ_F_MQROOT 8
53 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
54 * q->dev_queue : It can test
55 * netif_xmit_frozen_or_stopped() before
56 * dequeueing next packet.
57 * Its true for MQ/MQPRIO slaves, or non
60 #define TCQ_F_WARN_NONWC (1 << 16)
62 const struct Qdisc_ops
*ops
;
63 struct qdisc_size_table __rcu
*stab
;
64 struct list_head list
;
67 int (*reshape_fail
)(struct sk_buff
*skb
,
72 /* This field is deprecated, but it is still used by CBQ
73 * and it will live until better solution will be invented.
75 struct Qdisc
*__parent
;
76 struct netdev_queue
*dev_queue
;
78 struct gnet_stats_rate_est64 rate_est
;
79 struct Qdisc
*next_sched
;
80 struct sk_buff
*gso_skb
;
82 * For performance sake on SMP, we put highly modified fields at the end
85 struct sk_buff_head q
;
86 struct gnet_stats_basic_packed bstats
;
88 struct gnet_stats_queue qstats
;
89 struct rcu_head rcu_head
;
93 spinlock_t busylock ____cacheline_aligned_in_smp
;
96 static inline bool qdisc_is_running(const struct Qdisc
*qdisc
)
98 return (qdisc
->__state
& __QDISC___STATE_RUNNING
) ? true : false;
101 static inline bool qdisc_run_begin(struct Qdisc
*qdisc
)
103 if (qdisc_is_running(qdisc
))
105 qdisc
->__state
|= __QDISC___STATE_RUNNING
;
109 static inline void qdisc_run_end(struct Qdisc
*qdisc
)
111 qdisc
->__state
&= ~__QDISC___STATE_RUNNING
;
114 static inline bool qdisc_is_throttled(const struct Qdisc
*qdisc
)
116 return test_bit(__QDISC_STATE_THROTTLED
, &qdisc
->state
) ? true : false;
119 static inline void qdisc_throttled(struct Qdisc
*qdisc
)
121 set_bit(__QDISC_STATE_THROTTLED
, &qdisc
->state
);
124 static inline void qdisc_unthrottled(struct Qdisc
*qdisc
)
126 clear_bit(__QDISC_STATE_THROTTLED
, &qdisc
->state
);
129 struct Qdisc_class_ops
{
130 /* Child qdisc manipulation */
131 struct netdev_queue
* (*select_queue
)(struct Qdisc
*, struct tcmsg
*);
132 int (*graft
)(struct Qdisc
*, unsigned long cl
,
133 struct Qdisc
*, struct Qdisc
**);
134 struct Qdisc
* (*leaf
)(struct Qdisc
*, unsigned long cl
);
135 void (*qlen_notify
)(struct Qdisc
*, unsigned long);
137 /* Class manipulation routines */
138 unsigned long (*get
)(struct Qdisc
*, u32 classid
);
139 void (*put
)(struct Qdisc
*, unsigned long);
140 int (*change
)(struct Qdisc
*, u32
, u32
,
141 struct nlattr
**, unsigned long *);
142 int (*delete)(struct Qdisc
*, unsigned long);
143 void (*walk
)(struct Qdisc
*, struct qdisc_walker
* arg
);
145 /* Filter manipulation */
146 struct tcf_proto
** (*tcf_chain
)(struct Qdisc
*, unsigned long);
147 unsigned long (*bind_tcf
)(struct Qdisc
*, unsigned long,
149 void (*unbind_tcf
)(struct Qdisc
*, unsigned long);
151 /* rtnetlink specific */
152 int (*dump
)(struct Qdisc
*, unsigned long,
153 struct sk_buff
*skb
, struct tcmsg
*);
154 int (*dump_stats
)(struct Qdisc
*, unsigned long,
159 struct Qdisc_ops
*next
;
160 const struct Qdisc_class_ops
*cl_ops
;
164 int (*enqueue
)(struct sk_buff
*, struct Qdisc
*);
165 struct sk_buff
* (*dequeue
)(struct Qdisc
*);
166 struct sk_buff
* (*peek
)(struct Qdisc
*);
167 unsigned int (*drop
)(struct Qdisc
*);
169 int (*init
)(struct Qdisc
*, struct nlattr
*arg
);
170 void (*reset
)(struct Qdisc
*);
171 void (*destroy
)(struct Qdisc
*);
172 int (*change
)(struct Qdisc
*, struct nlattr
*arg
);
173 void (*attach
)(struct Qdisc
*);
175 int (*dump
)(struct Qdisc
*, struct sk_buff
*);
176 int (*dump_stats
)(struct Qdisc
*, struct gnet_dump
*);
178 struct module
*owner
;
187 struct tcf_proto_ops
{
188 struct tcf_proto_ops
*next
;
191 int (*classify
)(struct sk_buff
*,
192 const struct tcf_proto
*,
193 struct tcf_result
*);
194 int (*init
)(struct tcf_proto
*);
195 void (*destroy
)(struct tcf_proto
*);
197 unsigned long (*get
)(struct tcf_proto
*, u32 handle
);
198 void (*put
)(struct tcf_proto
*, unsigned long);
199 int (*change
)(struct net
*net
, struct sk_buff
*,
200 struct tcf_proto
*, unsigned long,
201 u32 handle
, struct nlattr
**,
203 int (*delete)(struct tcf_proto
*, unsigned long);
204 void (*walk
)(struct tcf_proto
*, struct tcf_walker
*arg
);
206 /* rtnetlink specific */
207 int (*dump
)(struct tcf_proto
*, unsigned long,
208 struct sk_buff
*skb
, struct tcmsg
*);
210 struct module
*owner
;
214 /* Fast access part */
215 struct tcf_proto
*next
;
217 int (*classify
)(struct sk_buff
*,
218 const struct tcf_proto
*,
219 struct tcf_result
*);
227 const struct tcf_proto_ops
*ops
;
230 struct qdisc_skb_cb
{
231 unsigned int pkt_len
;
232 u16 slave_dev_queue_mapping
;
234 unsigned char data
[20];
237 static inline void qdisc_cb_private_validate(const struct sk_buff
*skb
, int sz
)
239 struct qdisc_skb_cb
*qcb
;
241 BUILD_BUG_ON(sizeof(skb
->cb
) < offsetof(struct qdisc_skb_cb
, data
) + sz
);
242 BUILD_BUG_ON(sizeof(qcb
->data
) < sz
);
245 static inline int qdisc_qlen(const struct Qdisc
*q
)
250 static inline struct qdisc_skb_cb
*qdisc_skb_cb(const struct sk_buff
*skb
)
252 return (struct qdisc_skb_cb
*)skb
->cb
;
255 static inline spinlock_t
*qdisc_lock(struct Qdisc
*qdisc
)
257 return &qdisc
->q
.lock
;
260 static inline struct Qdisc
*qdisc_root(const struct Qdisc
*qdisc
)
262 return qdisc
->dev_queue
->qdisc
;
265 static inline struct Qdisc
*qdisc_root_sleeping(const struct Qdisc
*qdisc
)
267 return qdisc
->dev_queue
->qdisc_sleeping
;
270 /* The qdisc root lock is a mechanism by which to top level
271 * of a qdisc tree can be locked from any qdisc node in the
272 * forest. This allows changing the configuration of some
273 * aspect of the qdisc tree while blocking out asynchronous
274 * qdisc access in the packet processing paths.
276 * It is only legal to do this when the root will not change
277 * on us. Otherwise we'll potentially lock the wrong qdisc
278 * root. This is enforced by holding the RTNL semaphore, which
279 * all users of this lock accessor must do.
281 static inline spinlock_t
*qdisc_root_lock(const struct Qdisc
*qdisc
)
283 struct Qdisc
*root
= qdisc_root(qdisc
);
286 return qdisc_lock(root
);
289 static inline spinlock_t
*qdisc_root_sleeping_lock(const struct Qdisc
*qdisc
)
291 struct Qdisc
*root
= qdisc_root_sleeping(qdisc
);
294 return qdisc_lock(root
);
297 static inline struct net_device
*qdisc_dev(const struct Qdisc
*qdisc
)
299 return qdisc
->dev_queue
->dev
;
302 static inline void sch_tree_lock(const struct Qdisc
*q
)
304 spin_lock_bh(qdisc_root_sleeping_lock(q
));
307 static inline void sch_tree_unlock(const struct Qdisc
*q
)
309 spin_unlock_bh(qdisc_root_sleeping_lock(q
));
312 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
313 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
315 extern struct Qdisc noop_qdisc
;
316 extern struct Qdisc_ops noop_qdisc_ops
;
317 extern struct Qdisc_ops pfifo_fast_ops
;
318 extern struct Qdisc_ops mq_qdisc_ops
;
319 extern const struct Qdisc_ops
*default_qdisc_ops
;
321 struct Qdisc_class_common
{
323 struct hlist_node hnode
;
326 struct Qdisc_class_hash
{
327 struct hlist_head
*hash
;
328 unsigned int hashsize
;
329 unsigned int hashmask
;
330 unsigned int hashelems
;
333 static inline unsigned int qdisc_class_hash(u32 id
, u32 mask
)
340 static inline struct Qdisc_class_common
*
341 qdisc_class_find(const struct Qdisc_class_hash
*hash
, u32 id
)
343 struct Qdisc_class_common
*cl
;
346 h
= qdisc_class_hash(id
, hash
->hashmask
);
347 hlist_for_each_entry(cl
, &hash
->hash
[h
], hnode
) {
348 if (cl
->classid
== id
)
354 int qdisc_class_hash_init(struct Qdisc_class_hash
*);
355 void qdisc_class_hash_insert(struct Qdisc_class_hash
*,
356 struct Qdisc_class_common
*);
357 void qdisc_class_hash_remove(struct Qdisc_class_hash
*,
358 struct Qdisc_class_common
*);
359 void qdisc_class_hash_grow(struct Qdisc
*, struct Qdisc_class_hash
*);
360 void qdisc_class_hash_destroy(struct Qdisc_class_hash
*);
362 void dev_init_scheduler(struct net_device
*dev
);
363 void dev_shutdown(struct net_device
*dev
);
364 void dev_activate(struct net_device
*dev
);
365 void dev_deactivate(struct net_device
*dev
);
366 void dev_deactivate_many(struct list_head
*head
);
367 struct Qdisc
*dev_graft_qdisc(struct netdev_queue
*dev_queue
,
368 struct Qdisc
*qdisc
);
369 void qdisc_reset(struct Qdisc
*qdisc
);
370 void qdisc_destroy(struct Qdisc
*qdisc
);
371 void qdisc_tree_decrease_qlen(struct Qdisc
*qdisc
, unsigned int n
);
372 struct Qdisc
*qdisc_alloc(struct netdev_queue
*dev_queue
,
373 const struct Qdisc_ops
*ops
);
374 struct Qdisc
*qdisc_create_dflt(struct netdev_queue
*dev_queue
,
375 const struct Qdisc_ops
*ops
, u32 parentid
);
376 void __qdisc_calculate_pkt_len(struct sk_buff
*skb
,
377 const struct qdisc_size_table
*stab
);
378 void tcf_destroy(struct tcf_proto
*tp
);
379 void tcf_destroy_chain(struct tcf_proto
**fl
);
381 /* Reset all TX qdiscs greater then index of a device. */
382 static inline void qdisc_reset_all_tx_gt(struct net_device
*dev
, unsigned int i
)
386 for (; i
< dev
->num_tx_queues
; i
++) {
387 qdisc
= netdev_get_tx_queue(dev
, i
)->qdisc
;
389 spin_lock_bh(qdisc_lock(qdisc
));
391 spin_unlock_bh(qdisc_lock(qdisc
));
396 static inline void qdisc_reset_all_tx(struct net_device
*dev
)
398 qdisc_reset_all_tx_gt(dev
, 0);
401 /* Are all TX queues of the device empty? */
402 static inline bool qdisc_all_tx_empty(const struct net_device
*dev
)
405 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
406 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
407 const struct Qdisc
*q
= txq
->qdisc
;
415 /* Are any of the TX qdiscs changing? */
416 static inline bool qdisc_tx_changing(const struct net_device
*dev
)
419 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
420 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
421 if (txq
->qdisc
!= txq
->qdisc_sleeping
)
427 /* Is the device using the noop qdisc on all queues? */
428 static inline bool qdisc_tx_is_noop(const struct net_device
*dev
)
431 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
432 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
433 if (txq
->qdisc
!= &noop_qdisc
)
439 static inline unsigned int qdisc_pkt_len(const struct sk_buff
*skb
)
441 return qdisc_skb_cb(skb
)->pkt_len
;
444 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
445 enum net_xmit_qdisc_t
{
446 __NET_XMIT_STOLEN
= 0x00010000,
447 __NET_XMIT_BYPASS
= 0x00020000,
450 #ifdef CONFIG_NET_CLS_ACT
451 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
453 #define net_xmit_drop_count(e) (1)
456 static inline void qdisc_calculate_pkt_len(struct sk_buff
*skb
,
457 const struct Qdisc
*sch
)
459 #ifdef CONFIG_NET_SCHED
460 struct qdisc_size_table
*stab
= rcu_dereference_bh(sch
->stab
);
463 __qdisc_calculate_pkt_len(skb
, stab
);
467 static inline int qdisc_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
)
469 qdisc_calculate_pkt_len(skb
, sch
);
470 return sch
->enqueue(skb
, sch
);
473 static inline int qdisc_enqueue_root(struct sk_buff
*skb
, struct Qdisc
*sch
)
475 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
476 return qdisc_enqueue(skb
, sch
) & NET_XMIT_MASK
;
480 static inline void bstats_update(struct gnet_stats_basic_packed
*bstats
,
481 const struct sk_buff
*skb
)
483 bstats
->bytes
+= qdisc_pkt_len(skb
);
484 bstats
->packets
+= skb_is_gso(skb
) ? skb_shinfo(skb
)->gso_segs
: 1;
487 static inline void qdisc_bstats_update(struct Qdisc
*sch
,
488 const struct sk_buff
*skb
)
490 bstats_update(&sch
->bstats
, skb
);
493 static inline int __qdisc_enqueue_tail(struct sk_buff
*skb
, struct Qdisc
*sch
,
494 struct sk_buff_head
*list
)
496 __skb_queue_tail(list
, skb
);
497 sch
->qstats
.backlog
+= qdisc_pkt_len(skb
);
499 return NET_XMIT_SUCCESS
;
502 static inline int qdisc_enqueue_tail(struct sk_buff
*skb
, struct Qdisc
*sch
)
504 return __qdisc_enqueue_tail(skb
, sch
, &sch
->q
);
507 static inline struct sk_buff
*__qdisc_dequeue_head(struct Qdisc
*sch
,
508 struct sk_buff_head
*list
)
510 struct sk_buff
*skb
= __skb_dequeue(list
);
512 if (likely(skb
!= NULL
)) {
513 sch
->qstats
.backlog
-= qdisc_pkt_len(skb
);
514 qdisc_bstats_update(sch
, skb
);
520 static inline struct sk_buff
*qdisc_dequeue_head(struct Qdisc
*sch
)
522 return __qdisc_dequeue_head(sch
, &sch
->q
);
525 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc
*sch
,
526 struct sk_buff_head
*list
)
528 struct sk_buff
*skb
= __skb_dequeue(list
);
530 if (likely(skb
!= NULL
)) {
531 unsigned int len
= qdisc_pkt_len(skb
);
532 sch
->qstats
.backlog
-= len
;
540 static inline unsigned int qdisc_queue_drop_head(struct Qdisc
*sch
)
542 return __qdisc_queue_drop_head(sch
, &sch
->q
);
545 static inline struct sk_buff
*__qdisc_dequeue_tail(struct Qdisc
*sch
,
546 struct sk_buff_head
*list
)
548 struct sk_buff
*skb
= __skb_dequeue_tail(list
);
550 if (likely(skb
!= NULL
))
551 sch
->qstats
.backlog
-= qdisc_pkt_len(skb
);
556 static inline struct sk_buff
*qdisc_dequeue_tail(struct Qdisc
*sch
)
558 return __qdisc_dequeue_tail(sch
, &sch
->q
);
561 static inline struct sk_buff
*qdisc_peek_head(struct Qdisc
*sch
)
563 return skb_peek(&sch
->q
);
566 /* generic pseudo peek method for non-work-conserving qdisc */
567 static inline struct sk_buff
*qdisc_peek_dequeued(struct Qdisc
*sch
)
569 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
571 sch
->gso_skb
= sch
->dequeue(sch
);
573 /* it's still part of the queue */
580 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
581 static inline struct sk_buff
*qdisc_dequeue_peeked(struct Qdisc
*sch
)
583 struct sk_buff
*skb
= sch
->gso_skb
;
589 skb
= sch
->dequeue(sch
);
595 static inline void __qdisc_reset_queue(struct Qdisc
*sch
,
596 struct sk_buff_head
*list
)
599 * We do not know the backlog in bytes of this list, it
600 * is up to the caller to correct it
602 __skb_queue_purge(list
);
605 static inline void qdisc_reset_queue(struct Qdisc
*sch
)
607 __qdisc_reset_queue(sch
, &sch
->q
);
608 sch
->qstats
.backlog
= 0;
611 static inline unsigned int __qdisc_queue_drop(struct Qdisc
*sch
,
612 struct sk_buff_head
*list
)
614 struct sk_buff
*skb
= __qdisc_dequeue_tail(sch
, list
);
616 if (likely(skb
!= NULL
)) {
617 unsigned int len
= qdisc_pkt_len(skb
);
625 static inline unsigned int qdisc_queue_drop(struct Qdisc
*sch
)
627 return __qdisc_queue_drop(sch
, &sch
->q
);
630 static inline int qdisc_drop(struct sk_buff
*skb
, struct Qdisc
*sch
)
635 return NET_XMIT_DROP
;
638 static inline int qdisc_reshape_fail(struct sk_buff
*skb
, struct Qdisc
*sch
)
642 #ifdef CONFIG_NET_CLS_ACT
643 if (sch
->reshape_fail
== NULL
|| sch
->reshape_fail(skb
, sch
))
646 return NET_XMIT_SUCCESS
;
651 return NET_XMIT_DROP
;
654 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
655 long it will take to send a packet given its size.
657 static inline u32
qdisc_l2t(struct qdisc_rate_table
* rtab
, unsigned int pktlen
)
659 int slot
= pktlen
+ rtab
->rate
.cell_align
+ rtab
->rate
.overhead
;
662 slot
>>= rtab
->rate
.cell_log
;
664 return rtab
->data
[255]*(slot
>> 8) + rtab
->data
[slot
& 0xFF];
665 return rtab
->data
[slot
];
668 #ifdef CONFIG_NET_CLS_ACT
669 static inline struct sk_buff
*skb_act_clone(struct sk_buff
*skb
, gfp_t gfp_mask
,
674 n
= skb_clone(skb
, gfp_mask
);
677 n
->tc_verd
= SET_TC_VERD(n
->tc_verd
, 0);
678 n
->tc_verd
= CLR_TC_OK2MUNGE(n
->tc_verd
);
679 n
->tc_verd
= CLR_TC_MUNGED(n
->tc_verd
);
685 struct psched_ratecfg
{
686 u64 rate_bytes_ps
; /* bytes per second */
693 static inline u64
psched_l2t_ns(const struct psched_ratecfg
*r
,
698 if (unlikely(r
->linklayer
== TC_LINKLAYER_ATM
))
699 return ((u64
)(DIV_ROUND_UP(len
,48)*53) * r
->mult
) >> r
->shift
;
701 return ((u64
)len
* r
->mult
) >> r
->shift
;
704 void psched_ratecfg_precompute(struct psched_ratecfg
*r
,
705 const struct tc_ratespec
*conf
);
707 static inline void psched_ratecfg_getrate(struct tc_ratespec
*res
,
708 const struct psched_ratecfg
*r
)
710 memset(res
, 0, sizeof(*res
));
711 res
->rate
= r
->rate_bytes_ps
;
712 res
->overhead
= r
->overhead
;
713 res
->linklayer
= (r
->linklayer
& TC_LINKLAYER_MASK
);