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_WARN_NONWC (1 << 16)
55 const struct Qdisc_ops
*ops
;
56 struct qdisc_size_table __rcu
*stab
;
57 struct list_head list
;
61 struct gnet_stats_rate_est rate_est
;
62 int (*reshape_fail
)(struct sk_buff
*skb
,
67 /* This field is deprecated, but it is still used by CBQ
68 * and it will live until better solution will be invented.
70 struct Qdisc
*__parent
;
71 struct netdev_queue
*dev_queue
;
72 struct Qdisc
*next_sched
;
74 struct sk_buff
*gso_skb
;
76 * For performance sake on SMP, we put highly modified fields at the end
79 struct sk_buff_head q
;
80 struct gnet_stats_basic_packed bstats
;
82 struct gnet_stats_queue qstats
;
83 struct rcu_head rcu_head
;
88 static inline bool qdisc_is_running(const struct Qdisc
*qdisc
)
90 return (qdisc
->__state
& __QDISC___STATE_RUNNING
) ? true : false;
93 static inline bool qdisc_run_begin(struct Qdisc
*qdisc
)
95 if (qdisc_is_running(qdisc
))
97 qdisc
->__state
|= __QDISC___STATE_RUNNING
;
101 static inline void qdisc_run_end(struct Qdisc
*qdisc
)
103 qdisc
->__state
&= ~__QDISC___STATE_RUNNING
;
106 static inline bool qdisc_is_throttled(const struct Qdisc
*qdisc
)
108 return test_bit(__QDISC_STATE_THROTTLED
, &qdisc
->state
) ? true : false;
111 static inline void qdisc_throttled(struct Qdisc
*qdisc
)
113 set_bit(__QDISC_STATE_THROTTLED
, &qdisc
->state
);
116 static inline void qdisc_unthrottled(struct Qdisc
*qdisc
)
118 clear_bit(__QDISC_STATE_THROTTLED
, &qdisc
->state
);
121 struct Qdisc_class_ops
{
122 /* Child qdisc manipulation */
123 struct netdev_queue
* (*select_queue
)(struct Qdisc
*, struct tcmsg
*);
124 int (*graft
)(struct Qdisc
*, unsigned long cl
,
125 struct Qdisc
*, struct Qdisc
**);
126 struct Qdisc
* (*leaf
)(struct Qdisc
*, unsigned long cl
);
127 void (*qlen_notify
)(struct Qdisc
*, unsigned long);
129 /* Class manipulation routines */
130 unsigned long (*get
)(struct Qdisc
*, u32 classid
);
131 void (*put
)(struct Qdisc
*, unsigned long);
132 int (*change
)(struct Qdisc
*, u32
, u32
,
133 struct nlattr
**, unsigned long *);
134 int (*delete)(struct Qdisc
*, unsigned long);
135 void (*walk
)(struct Qdisc
*, struct qdisc_walker
* arg
);
137 /* Filter manipulation */
138 struct tcf_proto
** (*tcf_chain
)(struct Qdisc
*, unsigned long);
139 unsigned long (*bind_tcf
)(struct Qdisc
*, unsigned long,
141 void (*unbind_tcf
)(struct Qdisc
*, unsigned long);
143 /* rtnetlink specific */
144 int (*dump
)(struct Qdisc
*, unsigned long,
145 struct sk_buff
*skb
, struct tcmsg
*);
146 int (*dump_stats
)(struct Qdisc
*, unsigned long,
151 struct Qdisc_ops
*next
;
152 const struct Qdisc_class_ops
*cl_ops
;
156 int (*enqueue
)(struct sk_buff
*, struct Qdisc
*);
157 struct sk_buff
* (*dequeue
)(struct Qdisc
*);
158 struct sk_buff
* (*peek
)(struct Qdisc
*);
159 unsigned int (*drop
)(struct Qdisc
*);
161 int (*init
)(struct Qdisc
*, struct nlattr
*arg
);
162 void (*reset
)(struct Qdisc
*);
163 void (*destroy
)(struct Qdisc
*);
164 int (*change
)(struct Qdisc
*, struct nlattr
*arg
);
165 void (*attach
)(struct Qdisc
*);
167 int (*dump
)(struct Qdisc
*, struct sk_buff
*);
168 int (*dump_stats
)(struct Qdisc
*, struct gnet_dump
*);
170 struct module
*owner
;
179 struct tcf_proto_ops
{
180 struct tcf_proto_ops
*next
;
183 int (*classify
)(struct sk_buff
*,
184 const struct tcf_proto
*,
185 struct tcf_result
*);
186 int (*init
)(struct tcf_proto
*);
187 void (*destroy
)(struct tcf_proto
*);
189 unsigned long (*get
)(struct tcf_proto
*, u32 handle
);
190 void (*put
)(struct tcf_proto
*, unsigned long);
191 int (*change
)(struct tcf_proto
*, unsigned long,
192 u32 handle
, struct nlattr
**,
194 int (*delete)(struct tcf_proto
*, unsigned long);
195 void (*walk
)(struct tcf_proto
*, struct tcf_walker
*arg
);
197 /* rtnetlink specific */
198 int (*dump
)(struct tcf_proto
*, unsigned long,
199 struct sk_buff
*skb
, struct tcmsg
*);
201 struct module
*owner
;
205 /* Fast access part */
206 struct tcf_proto
*next
;
208 int (*classify
)(struct sk_buff
*,
209 const struct tcf_proto
*,
210 struct tcf_result
*);
218 const struct tcf_proto_ops
*ops
;
221 struct qdisc_skb_cb
{
222 unsigned int pkt_len
;
223 u16 slave_dev_queue_mapping
;
225 unsigned char data
[20];
228 static inline void qdisc_cb_private_validate(const struct sk_buff
*skb
, int sz
)
230 struct qdisc_skb_cb
*qcb
;
232 BUILD_BUG_ON(sizeof(skb
->cb
) < offsetof(struct qdisc_skb_cb
, data
) + sz
);
233 BUILD_BUG_ON(sizeof(qcb
->data
) < sz
);
236 static inline int qdisc_qlen(const struct Qdisc
*q
)
241 static inline struct qdisc_skb_cb
*qdisc_skb_cb(const struct sk_buff
*skb
)
243 return (struct qdisc_skb_cb
*)skb
->cb
;
246 static inline spinlock_t
*qdisc_lock(struct Qdisc
*qdisc
)
248 return &qdisc
->q
.lock
;
251 static inline struct Qdisc
*qdisc_root(const struct Qdisc
*qdisc
)
253 return qdisc
->dev_queue
->qdisc
;
256 static inline struct Qdisc
*qdisc_root_sleeping(const struct Qdisc
*qdisc
)
258 return qdisc
->dev_queue
->qdisc_sleeping
;
261 /* The qdisc root lock is a mechanism by which to top level
262 * of a qdisc tree can be locked from any qdisc node in the
263 * forest. This allows changing the configuration of some
264 * aspect of the qdisc tree while blocking out asynchronous
265 * qdisc access in the packet processing paths.
267 * It is only legal to do this when the root will not change
268 * on us. Otherwise we'll potentially lock the wrong qdisc
269 * root. This is enforced by holding the RTNL semaphore, which
270 * all users of this lock accessor must do.
272 static inline spinlock_t
*qdisc_root_lock(const struct Qdisc
*qdisc
)
274 struct Qdisc
*root
= qdisc_root(qdisc
);
277 return qdisc_lock(root
);
280 static inline spinlock_t
*qdisc_root_sleeping_lock(const struct Qdisc
*qdisc
)
282 struct Qdisc
*root
= qdisc_root_sleeping(qdisc
);
285 return qdisc_lock(root
);
288 static inline struct net_device
*qdisc_dev(const struct Qdisc
*qdisc
)
290 return qdisc
->dev_queue
->dev
;
293 static inline void sch_tree_lock(const struct Qdisc
*q
)
295 spin_lock_bh(qdisc_root_sleeping_lock(q
));
298 static inline void sch_tree_unlock(const struct Qdisc
*q
)
300 spin_unlock_bh(qdisc_root_sleeping_lock(q
));
303 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q)
304 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q)
306 extern struct Qdisc noop_qdisc
;
307 extern struct Qdisc_ops noop_qdisc_ops
;
308 extern struct Qdisc_ops pfifo_fast_ops
;
309 extern struct Qdisc_ops mq_qdisc_ops
;
311 struct Qdisc_class_common
{
313 struct hlist_node hnode
;
316 struct Qdisc_class_hash
{
317 struct hlist_head
*hash
;
318 unsigned int hashsize
;
319 unsigned int hashmask
;
320 unsigned int hashelems
;
323 static inline unsigned int qdisc_class_hash(u32 id
, u32 mask
)
330 static inline struct Qdisc_class_common
*
331 qdisc_class_find(const struct Qdisc_class_hash
*hash
, u32 id
)
333 struct Qdisc_class_common
*cl
;
334 struct hlist_node
*n
;
337 h
= qdisc_class_hash(id
, hash
->hashmask
);
338 hlist_for_each_entry(cl
, n
, &hash
->hash
[h
], hnode
) {
339 if (cl
->classid
== id
)
345 extern int qdisc_class_hash_init(struct Qdisc_class_hash
*);
346 extern void qdisc_class_hash_insert(struct Qdisc_class_hash
*, struct Qdisc_class_common
*);
347 extern void qdisc_class_hash_remove(struct Qdisc_class_hash
*, struct Qdisc_class_common
*);
348 extern void qdisc_class_hash_grow(struct Qdisc
*, struct Qdisc_class_hash
*);
349 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash
*);
351 extern void dev_init_scheduler(struct net_device
*dev
);
352 extern void dev_shutdown(struct net_device
*dev
);
353 extern void dev_activate(struct net_device
*dev
);
354 extern void dev_deactivate(struct net_device
*dev
);
355 extern void dev_deactivate_many(struct list_head
*head
);
356 extern struct Qdisc
*dev_graft_qdisc(struct netdev_queue
*dev_queue
,
357 struct Qdisc
*qdisc
);
358 extern void qdisc_reset(struct Qdisc
*qdisc
);
359 extern void qdisc_destroy(struct Qdisc
*qdisc
);
360 extern void qdisc_tree_decrease_qlen(struct Qdisc
*qdisc
, unsigned int n
);
361 extern struct Qdisc
*qdisc_alloc(struct netdev_queue
*dev_queue
,
362 struct Qdisc_ops
*ops
);
363 extern struct Qdisc
*qdisc_create_dflt(struct netdev_queue
*dev_queue
,
364 struct Qdisc_ops
*ops
, u32 parentid
);
365 extern void __qdisc_calculate_pkt_len(struct sk_buff
*skb
,
366 const struct qdisc_size_table
*stab
);
367 extern void tcf_destroy(struct tcf_proto
*tp
);
368 extern void tcf_destroy_chain(struct tcf_proto
**fl
);
370 /* Reset all TX qdiscs greater then index of a device. */
371 static inline void qdisc_reset_all_tx_gt(struct net_device
*dev
, unsigned int i
)
375 for (; i
< dev
->num_tx_queues
; i
++) {
376 qdisc
= netdev_get_tx_queue(dev
, i
)->qdisc
;
378 spin_lock_bh(qdisc_lock(qdisc
));
380 spin_unlock_bh(qdisc_lock(qdisc
));
385 static inline void qdisc_reset_all_tx(struct net_device
*dev
)
387 qdisc_reset_all_tx_gt(dev
, 0);
390 /* Are all TX queues of the device empty? */
391 static inline bool qdisc_all_tx_empty(const struct net_device
*dev
)
394 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
395 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
396 const struct Qdisc
*q
= txq
->qdisc
;
404 /* Are any of the TX qdiscs changing? */
405 static inline bool qdisc_tx_changing(const struct net_device
*dev
)
408 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
409 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
410 if (txq
->qdisc
!= txq
->qdisc_sleeping
)
416 /* Is the device using the noop qdisc on all queues? */
417 static inline bool qdisc_tx_is_noop(const struct net_device
*dev
)
420 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
421 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
422 if (txq
->qdisc
!= &noop_qdisc
)
428 static inline unsigned int qdisc_pkt_len(const struct sk_buff
*skb
)
430 return qdisc_skb_cb(skb
)->pkt_len
;
433 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
434 enum net_xmit_qdisc_t
{
435 __NET_XMIT_STOLEN
= 0x00010000,
436 __NET_XMIT_BYPASS
= 0x00020000,
439 #ifdef CONFIG_NET_CLS_ACT
440 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
442 #define net_xmit_drop_count(e) (1)
445 static inline void qdisc_calculate_pkt_len(struct sk_buff
*skb
,
446 const struct Qdisc
*sch
)
448 #ifdef CONFIG_NET_SCHED
449 struct qdisc_size_table
*stab
= rcu_dereference_bh(sch
->stab
);
452 __qdisc_calculate_pkt_len(skb
, stab
);
456 static inline int qdisc_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
)
458 qdisc_calculate_pkt_len(skb
, sch
);
459 return sch
->enqueue(skb
, sch
);
462 static inline int qdisc_enqueue_root(struct sk_buff
*skb
, struct Qdisc
*sch
)
464 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
465 return qdisc_enqueue(skb
, sch
) & NET_XMIT_MASK
;
469 static inline void bstats_update(struct gnet_stats_basic_packed
*bstats
,
470 const struct sk_buff
*skb
)
472 bstats
->bytes
+= qdisc_pkt_len(skb
);
473 bstats
->packets
+= skb_is_gso(skb
) ? skb_shinfo(skb
)->gso_segs
: 1;
476 static inline void qdisc_bstats_update(struct Qdisc
*sch
,
477 const struct sk_buff
*skb
)
479 bstats_update(&sch
->bstats
, skb
);
482 static inline int __qdisc_enqueue_tail(struct sk_buff
*skb
, struct Qdisc
*sch
,
483 struct sk_buff_head
*list
)
485 __skb_queue_tail(list
, skb
);
486 sch
->qstats
.backlog
+= qdisc_pkt_len(skb
);
488 return NET_XMIT_SUCCESS
;
491 static inline int qdisc_enqueue_tail(struct sk_buff
*skb
, struct Qdisc
*sch
)
493 return __qdisc_enqueue_tail(skb
, sch
, &sch
->q
);
496 static inline struct sk_buff
*__qdisc_dequeue_head(struct Qdisc
*sch
,
497 struct sk_buff_head
*list
)
499 struct sk_buff
*skb
= __skb_dequeue(list
);
501 if (likely(skb
!= NULL
)) {
502 sch
->qstats
.backlog
-= qdisc_pkt_len(skb
);
503 qdisc_bstats_update(sch
, skb
);
509 static inline struct sk_buff
*qdisc_dequeue_head(struct Qdisc
*sch
)
511 return __qdisc_dequeue_head(sch
, &sch
->q
);
514 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc
*sch
,
515 struct sk_buff_head
*list
)
517 struct sk_buff
*skb
= __skb_dequeue(list
);
519 if (likely(skb
!= NULL
)) {
520 unsigned int len
= qdisc_pkt_len(skb
);
521 sch
->qstats
.backlog
-= len
;
529 static inline unsigned int qdisc_queue_drop_head(struct Qdisc
*sch
)
531 return __qdisc_queue_drop_head(sch
, &sch
->q
);
534 static inline struct sk_buff
*__qdisc_dequeue_tail(struct Qdisc
*sch
,
535 struct sk_buff_head
*list
)
537 struct sk_buff
*skb
= __skb_dequeue_tail(list
);
539 if (likely(skb
!= NULL
))
540 sch
->qstats
.backlog
-= qdisc_pkt_len(skb
);
545 static inline struct sk_buff
*qdisc_dequeue_tail(struct Qdisc
*sch
)
547 return __qdisc_dequeue_tail(sch
, &sch
->q
);
550 static inline struct sk_buff
*qdisc_peek_head(struct Qdisc
*sch
)
552 return skb_peek(&sch
->q
);
555 /* generic pseudo peek method for non-work-conserving qdisc */
556 static inline struct sk_buff
*qdisc_peek_dequeued(struct Qdisc
*sch
)
558 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
560 sch
->gso_skb
= sch
->dequeue(sch
);
562 /* it's still part of the queue */
569 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
570 static inline struct sk_buff
*qdisc_dequeue_peeked(struct Qdisc
*sch
)
572 struct sk_buff
*skb
= sch
->gso_skb
;
578 skb
= sch
->dequeue(sch
);
584 static inline void __qdisc_reset_queue(struct Qdisc
*sch
,
585 struct sk_buff_head
*list
)
588 * We do not know the backlog in bytes of this list, it
589 * is up to the caller to correct it
591 __skb_queue_purge(list
);
594 static inline void qdisc_reset_queue(struct Qdisc
*sch
)
596 __qdisc_reset_queue(sch
, &sch
->q
);
597 sch
->qstats
.backlog
= 0;
600 static inline unsigned int __qdisc_queue_drop(struct Qdisc
*sch
,
601 struct sk_buff_head
*list
)
603 struct sk_buff
*skb
= __qdisc_dequeue_tail(sch
, list
);
605 if (likely(skb
!= NULL
)) {
606 unsigned int len
= qdisc_pkt_len(skb
);
614 static inline unsigned int qdisc_queue_drop(struct Qdisc
*sch
)
616 return __qdisc_queue_drop(sch
, &sch
->q
);
619 static inline int qdisc_drop(struct sk_buff
*skb
, struct Qdisc
*sch
)
624 return NET_XMIT_DROP
;
627 static inline int qdisc_reshape_fail(struct sk_buff
*skb
, struct Qdisc
*sch
)
631 #ifdef CONFIG_NET_CLS_ACT
632 if (sch
->reshape_fail
== NULL
|| sch
->reshape_fail(skb
, sch
))
635 return NET_XMIT_SUCCESS
;
640 return NET_XMIT_DROP
;
643 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
644 long it will take to send a packet given its size.
646 static inline u32
qdisc_l2t(struct qdisc_rate_table
* rtab
, unsigned int pktlen
)
648 int slot
= pktlen
+ rtab
->rate
.cell_align
+ rtab
->rate
.overhead
;
651 slot
>>= rtab
->rate
.cell_log
;
653 return rtab
->data
[255]*(slot
>> 8) + rtab
->data
[slot
& 0xFF];
654 return rtab
->data
[slot
];
657 #ifdef CONFIG_NET_CLS_ACT
658 static inline struct sk_buff
*skb_act_clone(struct sk_buff
*skb
, gfp_t gfp_mask
,
663 n
= skb_clone(skb
, gfp_mask
);
666 n
->tc_verd
= SET_TC_VERD(n
->tc_verd
, 0);
667 n
->tc_verd
= CLR_TC_OK2MUNGE(n
->tc_verd
);
668 n
->tc_verd
= CLR_TC_MUNGED(n
->tc_verd
);