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 <linux/percpu.h>
10 #include <linux/dynamic_queue_limits.h>
11 #include <net/gen_stats.h>
12 #include <net/rtnetlink.h>
19 struct qdisc_rate_table
{
20 struct tc_ratespec rate
;
22 struct qdisc_rate_table
*next
;
28 __QDISC_STATE_DEACTIVATED
,
29 __QDISC_STATE_THROTTLED
,
33 * following bits are only changed while qdisc lock is held
35 enum qdisc___state_t
{
36 __QDISC___STATE_RUNNING
= 1,
39 struct qdisc_size_table
{
41 struct list_head list
;
42 struct tc_sizespec szopts
;
48 int (*enqueue
)(struct sk_buff
*skb
, struct Qdisc
*dev
);
49 struct sk_buff
* (*dequeue
)(struct Qdisc
*dev
);
51 #define TCQ_F_BUILTIN 1
52 #define TCQ_F_INGRESS 2
53 #define TCQ_F_CAN_BYPASS 4
54 #define TCQ_F_MQROOT 8
55 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
56 * q->dev_queue : It can test
57 * netif_xmit_frozen_or_stopped() before
58 * dequeueing next packet.
59 * Its true for MQ/MQPRIO slaves, or non
62 #define TCQ_F_WARN_NONWC (1 << 16)
63 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
65 const struct Qdisc_ops
*ops
;
66 struct qdisc_size_table __rcu
*stab
;
67 struct list_head list
;
70 int (*reshape_fail
)(struct sk_buff
*skb
,
75 /* This field is deprecated, but it is still used by CBQ
76 * and it will live until better solution will be invented.
78 struct Qdisc
*__parent
;
79 struct netdev_queue
*dev_queue
;
81 struct gnet_stats_rate_est64 rate_est
;
82 struct gnet_stats_basic_cpu __percpu
*cpu_bstats
;
83 struct gnet_stats_queue __percpu
*cpu_qstats
;
85 struct Qdisc
*next_sched
;
86 struct sk_buff
*gso_skb
;
88 * For performance sake on SMP, we put highly modified fields at the end
91 struct sk_buff_head q
;
92 struct gnet_stats_basic_packed bstats
;
94 struct gnet_stats_queue qstats
;
95 struct rcu_head rcu_head
;
99 spinlock_t busylock ____cacheline_aligned_in_smp
;
102 static inline bool qdisc_is_running(const struct Qdisc
*qdisc
)
104 return (qdisc
->__state
& __QDISC___STATE_RUNNING
) ? true : false;
107 static inline bool qdisc_run_begin(struct Qdisc
*qdisc
)
109 if (qdisc_is_running(qdisc
))
111 qdisc
->__state
|= __QDISC___STATE_RUNNING
;
115 static inline void qdisc_run_end(struct Qdisc
*qdisc
)
117 qdisc
->__state
&= ~__QDISC___STATE_RUNNING
;
120 static inline bool qdisc_may_bulk(const struct Qdisc
*qdisc
)
122 return qdisc
->flags
& TCQ_F_ONETXQUEUE
;
125 static inline int qdisc_avail_bulklimit(const struct netdev_queue
*txq
)
128 /* Non-BQL migrated drivers will return 0, too. */
129 return dql_avail(&txq
->dql
);
135 static inline bool qdisc_is_throttled(const struct Qdisc
*qdisc
)
137 return test_bit(__QDISC_STATE_THROTTLED
, &qdisc
->state
) ? true : false;
140 static inline void qdisc_throttled(struct Qdisc
*qdisc
)
142 set_bit(__QDISC_STATE_THROTTLED
, &qdisc
->state
);
145 static inline void qdisc_unthrottled(struct Qdisc
*qdisc
)
147 clear_bit(__QDISC_STATE_THROTTLED
, &qdisc
->state
);
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 Qdisc
* (*leaf
)(struct Qdisc
*, unsigned long cl
);
156 void (*qlen_notify
)(struct Qdisc
*, unsigned long);
158 /* Class manipulation routines */
159 unsigned long (*get
)(struct Qdisc
*, u32 classid
);
160 void (*put
)(struct Qdisc
*, unsigned long);
161 int (*change
)(struct Qdisc
*, u32
, u32
,
162 struct nlattr
**, unsigned long *);
163 int (*delete)(struct Qdisc
*, unsigned long);
164 void (*walk
)(struct Qdisc
*, struct qdisc_walker
* arg
);
166 /* Filter manipulation */
167 struct tcf_proto __rcu
** (*tcf_chain
)(struct Qdisc
*, unsigned long);
168 unsigned long (*bind_tcf
)(struct Qdisc
*, unsigned long,
170 void (*unbind_tcf
)(struct Qdisc
*, unsigned long);
172 /* rtnetlink specific */
173 int (*dump
)(struct Qdisc
*, unsigned long,
174 struct sk_buff
*skb
, struct tcmsg
*);
175 int (*dump_stats
)(struct Qdisc
*, unsigned long,
180 struct Qdisc_ops
*next
;
181 const struct Qdisc_class_ops
*cl_ops
;
185 int (*enqueue
)(struct sk_buff
*, struct Qdisc
*);
186 struct sk_buff
* (*dequeue
)(struct Qdisc
*);
187 struct sk_buff
* (*peek
)(struct Qdisc
*);
188 unsigned int (*drop
)(struct Qdisc
*);
190 int (*init
)(struct Qdisc
*, struct nlattr
*arg
);
191 void (*reset
)(struct Qdisc
*);
192 void (*destroy
)(struct Qdisc
*);
193 int (*change
)(struct Qdisc
*, struct nlattr
*arg
);
194 void (*attach
)(struct Qdisc
*);
196 int (*dump
)(struct Qdisc
*, struct sk_buff
*);
197 int (*dump_stats
)(struct Qdisc
*, struct gnet_dump
*);
199 struct module
*owner
;
208 struct tcf_proto_ops
{
209 struct list_head head
;
212 int (*classify
)(struct sk_buff
*,
213 const struct tcf_proto
*,
214 struct tcf_result
*);
215 int (*init
)(struct tcf_proto
*);
216 bool (*destroy
)(struct tcf_proto
*, bool);
218 unsigned long (*get
)(struct tcf_proto
*, u32 handle
);
219 int (*change
)(struct net
*net
, struct sk_buff
*,
220 struct tcf_proto
*, unsigned long,
221 u32 handle
, struct nlattr
**,
222 unsigned long *, bool);
223 int (*delete)(struct tcf_proto
*, unsigned long);
224 void (*walk
)(struct tcf_proto
*, struct tcf_walker
*arg
);
226 /* rtnetlink specific */
227 int (*dump
)(struct net
*, struct tcf_proto
*, unsigned long,
228 struct sk_buff
*skb
, struct tcmsg
*);
230 struct module
*owner
;
234 /* Fast access part */
235 struct tcf_proto __rcu
*next
;
237 int (*classify
)(struct sk_buff
*,
238 const struct tcf_proto
*,
239 struct tcf_result
*);
247 const struct tcf_proto_ops
*ops
;
251 struct qdisc_skb_cb
{
252 unsigned int pkt_len
;
253 u16 slave_dev_queue_mapping
;
255 #define QDISC_CB_PRIV_LEN 20
256 unsigned char data
[QDISC_CB_PRIV_LEN
];
259 static inline void qdisc_cb_private_validate(const struct sk_buff
*skb
, int sz
)
261 struct qdisc_skb_cb
*qcb
;
263 BUILD_BUG_ON(sizeof(skb
->cb
) < offsetof(struct qdisc_skb_cb
, data
) + sz
);
264 BUILD_BUG_ON(sizeof(qcb
->data
) < sz
);
267 static inline int qdisc_qlen(const struct Qdisc
*q
)
272 static inline struct qdisc_skb_cb
*qdisc_skb_cb(const struct sk_buff
*skb
)
274 return (struct qdisc_skb_cb
*)skb
->cb
;
277 static inline spinlock_t
*qdisc_lock(struct Qdisc
*qdisc
)
279 return &qdisc
->q
.lock
;
282 static inline struct Qdisc
*qdisc_root(const struct Qdisc
*qdisc
)
284 struct Qdisc
*q
= rcu_dereference_rtnl(qdisc
->dev_queue
->qdisc
);
289 static inline struct Qdisc
*qdisc_root_sleeping(const struct Qdisc
*qdisc
)
291 return qdisc
->dev_queue
->qdisc_sleeping
;
294 /* The qdisc root lock is a mechanism by which to top level
295 * of a qdisc tree can be locked from any qdisc node in the
296 * forest. This allows changing the configuration of some
297 * aspect of the qdisc tree while blocking out asynchronous
298 * qdisc access in the packet processing paths.
300 * It is only legal to do this when the root will not change
301 * on us. Otherwise we'll potentially lock the wrong qdisc
302 * root. This is enforced by holding the RTNL semaphore, which
303 * all users of this lock accessor must do.
305 static inline spinlock_t
*qdisc_root_lock(const struct Qdisc
*qdisc
)
307 struct Qdisc
*root
= qdisc_root(qdisc
);
310 return qdisc_lock(root
);
313 static inline spinlock_t
*qdisc_root_sleeping_lock(const struct Qdisc
*qdisc
)
315 struct Qdisc
*root
= qdisc_root_sleeping(qdisc
);
318 return qdisc_lock(root
);
321 static inline struct net_device
*qdisc_dev(const struct Qdisc
*qdisc
)
323 return qdisc
->dev_queue
->dev
;
326 static inline void sch_tree_lock(const struct Qdisc
*q
)
328 spin_lock_bh(qdisc_root_sleeping_lock(q
));
331 static inline void sch_tree_unlock(const struct Qdisc
*q
)
333 spin_unlock_bh(qdisc_root_sleeping_lock(q
));
336 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
337 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
339 extern struct Qdisc noop_qdisc
;
340 extern struct Qdisc_ops noop_qdisc_ops
;
341 extern struct Qdisc_ops pfifo_fast_ops
;
342 extern struct Qdisc_ops mq_qdisc_ops
;
343 extern const struct Qdisc_ops
*default_qdisc_ops
;
345 struct Qdisc_class_common
{
347 struct hlist_node hnode
;
350 struct Qdisc_class_hash
{
351 struct hlist_head
*hash
;
352 unsigned int hashsize
;
353 unsigned int hashmask
;
354 unsigned int hashelems
;
357 static inline unsigned int qdisc_class_hash(u32 id
, u32 mask
)
364 static inline struct Qdisc_class_common
*
365 qdisc_class_find(const struct Qdisc_class_hash
*hash
, u32 id
)
367 struct Qdisc_class_common
*cl
;
370 h
= qdisc_class_hash(id
, hash
->hashmask
);
371 hlist_for_each_entry(cl
, &hash
->hash
[h
], hnode
) {
372 if (cl
->classid
== id
)
378 int qdisc_class_hash_init(struct Qdisc_class_hash
*);
379 void qdisc_class_hash_insert(struct Qdisc_class_hash
*,
380 struct Qdisc_class_common
*);
381 void qdisc_class_hash_remove(struct Qdisc_class_hash
*,
382 struct Qdisc_class_common
*);
383 void qdisc_class_hash_grow(struct Qdisc
*, struct Qdisc_class_hash
*);
384 void qdisc_class_hash_destroy(struct Qdisc_class_hash
*);
386 void dev_init_scheduler(struct net_device
*dev
);
387 void dev_shutdown(struct net_device
*dev
);
388 void dev_activate(struct net_device
*dev
);
389 void dev_deactivate(struct net_device
*dev
);
390 void dev_deactivate_many(struct list_head
*head
);
391 struct Qdisc
*dev_graft_qdisc(struct netdev_queue
*dev_queue
,
392 struct Qdisc
*qdisc
);
393 void qdisc_reset(struct Qdisc
*qdisc
);
394 void qdisc_destroy(struct Qdisc
*qdisc
);
395 void qdisc_tree_decrease_qlen(struct Qdisc
*qdisc
, unsigned int n
);
396 struct Qdisc
*qdisc_alloc(struct netdev_queue
*dev_queue
,
397 const struct Qdisc_ops
*ops
);
398 struct Qdisc
*qdisc_create_dflt(struct netdev_queue
*dev_queue
,
399 const struct Qdisc_ops
*ops
, u32 parentid
);
400 void __qdisc_calculate_pkt_len(struct sk_buff
*skb
,
401 const struct qdisc_size_table
*stab
);
402 bool tcf_destroy(struct tcf_proto
*tp
, bool force
);
403 void tcf_destroy_chain(struct tcf_proto __rcu
**fl
);
405 /* Reset all TX qdiscs greater then index of a device. */
406 static inline void qdisc_reset_all_tx_gt(struct net_device
*dev
, unsigned int i
)
410 for (; i
< dev
->num_tx_queues
; i
++) {
411 qdisc
= rtnl_dereference(netdev_get_tx_queue(dev
, i
)->qdisc
);
413 spin_lock_bh(qdisc_lock(qdisc
));
415 spin_unlock_bh(qdisc_lock(qdisc
));
420 static inline void qdisc_reset_all_tx(struct net_device
*dev
)
422 qdisc_reset_all_tx_gt(dev
, 0);
425 /* Are all TX queues of the device empty? */
426 static inline bool qdisc_all_tx_empty(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 const struct Qdisc
*q
= rcu_dereference(txq
->qdisc
);
444 /* Are any of the TX qdiscs changing? */
445 static inline bool qdisc_tx_changing(const struct net_device
*dev
)
449 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
450 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
451 if (rcu_access_pointer(txq
->qdisc
) != txq
->qdisc_sleeping
)
457 /* Is the device using the noop qdisc on all queues? */
458 static inline bool qdisc_tx_is_noop(const struct net_device
*dev
)
462 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
463 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
464 if (rcu_access_pointer(txq
->qdisc
) != &noop_qdisc
)
470 static inline unsigned int qdisc_pkt_len(const struct sk_buff
*skb
)
472 return qdisc_skb_cb(skb
)->pkt_len
;
475 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
476 enum net_xmit_qdisc_t
{
477 __NET_XMIT_STOLEN
= 0x00010000,
478 __NET_XMIT_BYPASS
= 0x00020000,
481 #ifdef CONFIG_NET_CLS_ACT
482 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
484 #define net_xmit_drop_count(e) (1)
487 static inline void qdisc_calculate_pkt_len(struct sk_buff
*skb
,
488 const struct Qdisc
*sch
)
490 #ifdef CONFIG_NET_SCHED
491 struct qdisc_size_table
*stab
= rcu_dereference_bh(sch
->stab
);
494 __qdisc_calculate_pkt_len(skb
, stab
);
498 static inline int qdisc_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
)
500 qdisc_calculate_pkt_len(skb
, sch
);
501 return sch
->enqueue(skb
, sch
);
504 static inline bool qdisc_is_percpu_stats(const struct Qdisc
*q
)
506 return q
->flags
& TCQ_F_CPUSTATS
;
509 static inline void bstats_update(struct gnet_stats_basic_packed
*bstats
,
510 const struct sk_buff
*skb
)
512 bstats
->bytes
+= qdisc_pkt_len(skb
);
513 bstats
->packets
+= skb_is_gso(skb
) ? skb_shinfo(skb
)->gso_segs
: 1;
516 static inline void qdisc_bstats_update_cpu(struct Qdisc
*sch
,
517 const struct sk_buff
*skb
)
519 struct gnet_stats_basic_cpu
*bstats
=
520 this_cpu_ptr(sch
->cpu_bstats
);
522 u64_stats_update_begin(&bstats
->syncp
);
523 bstats_update(&bstats
->bstats
, skb
);
524 u64_stats_update_end(&bstats
->syncp
);
527 static inline void qdisc_bstats_update(struct Qdisc
*sch
,
528 const struct sk_buff
*skb
)
530 bstats_update(&sch
->bstats
, skb
);
533 static inline void qdisc_qstats_backlog_dec(struct Qdisc
*sch
,
534 const struct sk_buff
*skb
)
536 sch
->qstats
.backlog
-= qdisc_pkt_len(skb
);
539 static inline void qdisc_qstats_backlog_inc(struct Qdisc
*sch
,
540 const struct sk_buff
*skb
)
542 sch
->qstats
.backlog
+= qdisc_pkt_len(skb
);
545 static inline void __qdisc_qstats_drop(struct Qdisc
*sch
, int count
)
547 sch
->qstats
.drops
+= count
;
550 static inline void qdisc_qstats_drop(struct Qdisc
*sch
)
555 static inline void qdisc_qstats_drop_cpu(struct Qdisc
*sch
)
557 struct gnet_stats_queue
*qstats
= this_cpu_ptr(sch
->cpu_qstats
);
562 static inline void qdisc_qstats_overlimit(struct Qdisc
*sch
)
564 sch
->qstats
.overlimits
++;
567 static inline int __qdisc_enqueue_tail(struct sk_buff
*skb
, struct Qdisc
*sch
,
568 struct sk_buff_head
*list
)
570 __skb_queue_tail(list
, skb
);
571 qdisc_qstats_backlog_inc(sch
, skb
);
573 return NET_XMIT_SUCCESS
;
576 static inline int qdisc_enqueue_tail(struct sk_buff
*skb
, struct Qdisc
*sch
)
578 return __qdisc_enqueue_tail(skb
, sch
, &sch
->q
);
581 static inline struct sk_buff
*__qdisc_dequeue_head(struct Qdisc
*sch
,
582 struct sk_buff_head
*list
)
584 struct sk_buff
*skb
= __skb_dequeue(list
);
586 if (likely(skb
!= NULL
)) {
587 qdisc_qstats_backlog_dec(sch
, skb
);
588 qdisc_bstats_update(sch
, skb
);
594 static inline struct sk_buff
*qdisc_dequeue_head(struct Qdisc
*sch
)
596 return __qdisc_dequeue_head(sch
, &sch
->q
);
599 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc
*sch
,
600 struct sk_buff_head
*list
)
602 struct sk_buff
*skb
= __skb_dequeue(list
);
604 if (likely(skb
!= NULL
)) {
605 unsigned int len
= qdisc_pkt_len(skb
);
606 qdisc_qstats_backlog_dec(sch
, skb
);
614 static inline unsigned int qdisc_queue_drop_head(struct Qdisc
*sch
)
616 return __qdisc_queue_drop_head(sch
, &sch
->q
);
619 static inline struct sk_buff
*__qdisc_dequeue_tail(struct Qdisc
*sch
,
620 struct sk_buff_head
*list
)
622 struct sk_buff
*skb
= __skb_dequeue_tail(list
);
624 if (likely(skb
!= NULL
))
625 qdisc_qstats_backlog_dec(sch
, skb
);
630 static inline struct sk_buff
*qdisc_dequeue_tail(struct Qdisc
*sch
)
632 return __qdisc_dequeue_tail(sch
, &sch
->q
);
635 static inline struct sk_buff
*qdisc_peek_head(struct Qdisc
*sch
)
637 return skb_peek(&sch
->q
);
640 /* generic pseudo peek method for non-work-conserving qdisc */
641 static inline struct sk_buff
*qdisc_peek_dequeued(struct Qdisc
*sch
)
643 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
645 sch
->gso_skb
= sch
->dequeue(sch
);
647 /* it's still part of the queue */
654 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
655 static inline struct sk_buff
*qdisc_dequeue_peeked(struct Qdisc
*sch
)
657 struct sk_buff
*skb
= sch
->gso_skb
;
663 skb
= sch
->dequeue(sch
);
669 static inline void __qdisc_reset_queue(struct Qdisc
*sch
,
670 struct sk_buff_head
*list
)
673 * We do not know the backlog in bytes of this list, it
674 * is up to the caller to correct it
676 __skb_queue_purge(list
);
679 static inline void qdisc_reset_queue(struct Qdisc
*sch
)
681 __qdisc_reset_queue(sch
, &sch
->q
);
682 sch
->qstats
.backlog
= 0;
685 static inline unsigned int __qdisc_queue_drop(struct Qdisc
*sch
,
686 struct sk_buff_head
*list
)
688 struct sk_buff
*skb
= __qdisc_dequeue_tail(sch
, list
);
690 if (likely(skb
!= NULL
)) {
691 unsigned int len
= qdisc_pkt_len(skb
);
699 static inline unsigned int qdisc_queue_drop(struct Qdisc
*sch
)
701 return __qdisc_queue_drop(sch
, &sch
->q
);
704 static inline int qdisc_drop(struct sk_buff
*skb
, struct Qdisc
*sch
)
707 qdisc_qstats_drop(sch
);
709 return NET_XMIT_DROP
;
712 static inline int qdisc_reshape_fail(struct sk_buff
*skb
, struct Qdisc
*sch
)
714 qdisc_qstats_drop(sch
);
716 #ifdef CONFIG_NET_CLS_ACT
717 if (sch
->reshape_fail
== NULL
|| sch
->reshape_fail(skb
, sch
))
720 return NET_XMIT_SUCCESS
;
725 return NET_XMIT_DROP
;
728 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
729 long it will take to send a packet given its size.
731 static inline u32
qdisc_l2t(struct qdisc_rate_table
* rtab
, unsigned int pktlen
)
733 int slot
= pktlen
+ rtab
->rate
.cell_align
+ rtab
->rate
.overhead
;
736 slot
>>= rtab
->rate
.cell_log
;
738 return rtab
->data
[255]*(slot
>> 8) + rtab
->data
[slot
& 0xFF];
739 return rtab
->data
[slot
];
742 struct psched_ratecfg
{
743 u64 rate_bytes_ps
; /* bytes per second */
750 static inline u64
psched_l2t_ns(const struct psched_ratecfg
*r
,
755 if (unlikely(r
->linklayer
== TC_LINKLAYER_ATM
))
756 return ((u64
)(DIV_ROUND_UP(len
,48)*53) * r
->mult
) >> r
->shift
;
758 return ((u64
)len
* r
->mult
) >> r
->shift
;
761 void psched_ratecfg_precompute(struct psched_ratecfg
*r
,
762 const struct tc_ratespec
*conf
,
765 static inline void psched_ratecfg_getrate(struct tc_ratespec
*res
,
766 const struct psched_ratecfg
*r
)
768 memset(res
, 0, sizeof(*res
));
770 /* legacy struct tc_ratespec has a 32bit @rate field
771 * Qdisc using 64bit rate should add new attributes
772 * in order to maintain compatibility.
774 res
->rate
= min_t(u64
, r
->rate_bytes_ps
, ~0U);
776 res
->overhead
= r
->overhead
;
777 res
->linklayer
= (r
->linklayer
& TC_LINKLAYER_MASK
);