2 * net/sched/sch_sfq.c Stochastic Fairness Queueing discipline.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
12 #include <linux/module.h>
13 #include <linux/types.h>
14 #include <linux/kernel.h>
15 #include <linux/jiffies.h>
16 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/init.h>
20 #include <linux/skbuff.h>
21 #include <linux/jhash.h>
22 #include <linux/slab.h>
23 #include <linux/vmalloc.h>
24 #include <net/netlink.h>
25 #include <net/pkt_sched.h>
26 #include <net/flow_keys.h>
30 /* Stochastic Fairness Queuing algorithm.
31 =======================================
34 Paul E. McKenney "Stochastic Fairness Queuing",
35 IEEE INFOCOMM'90 Proceedings, San Francisco, 1990.
37 Paul E. McKenney "Stochastic Fairness Queuing",
38 "Interworking: Research and Experience", v.2, 1991, p.113-131.
42 M. Shreedhar and George Varghese "Efficient Fair
43 Queuing using Deficit Round Robin", Proc. SIGCOMM 95.
46 This is not the thing that is usually called (W)FQ nowadays.
47 It does not use any timestamp mechanism, but instead
48 processes queues in round-robin order.
52 - It is very cheap. Both CPU and memory requirements are minimal.
56 - "Stochastic" -> It is not 100% fair.
57 When hash collisions occur, several flows are considered as one.
59 - "Round-robin" -> It introduces larger delays than virtual clock
60 based schemes, and should not be used for isolating interactive
61 traffic from non-interactive. It means, that this scheduler
62 should be used as leaf of CBQ or P3, which put interactive traffic
63 to higher priority band.
65 We still need true WFQ for top level CSZ, but using WFQ
66 for the best effort traffic is absolutely pointless:
67 SFQ is superior for this purpose.
70 This implementation limits :
71 - maximal queue length per flow to 127 packets.
74 - number of hash buckets to 65536.
76 It is easy to increase these values, but not in flight. */
78 #define SFQ_MAX_DEPTH 127 /* max number of packets per flow */
79 #define SFQ_DEFAULT_FLOWS 128
80 #define SFQ_MAX_FLOWS (0x10000 - SFQ_MAX_DEPTH - 1) /* max number of flows */
81 #define SFQ_EMPTY_SLOT 0xffff
82 #define SFQ_DEFAULT_HASH_DIVISOR 1024
84 /* We use 16 bits to store allot, and want to handle packets up to 64K
85 * Scale allot by 8 (1<<3) so that no overflow occurs.
87 #define SFQ_ALLOT_SHIFT 3
88 #define SFQ_ALLOT_SIZE(X) DIV_ROUND_UP(X, 1 << SFQ_ALLOT_SHIFT)
90 /* This type should contain at least SFQ_MAX_DEPTH + 1 + SFQ_MAX_FLOWS values */
91 typedef u16 sfq_index
;
94 * We dont use pointers to save space.
95 * Small indexes [0 ... SFQ_MAX_FLOWS - 1] are 'pointers' to slots[] array
96 * while following values [SFQ_MAX_FLOWS ... SFQ_MAX_FLOWS + SFQ_MAX_DEPTH]
97 * are 'pointers' to dep[] array
105 struct sk_buff
*skblist_next
;
106 struct sk_buff
*skblist_prev
;
107 sfq_index qlen
; /* number of skbs in skblist */
108 sfq_index next
; /* next slot in sfq RR chain */
109 struct sfq_head dep
; /* anchor in dep[] chains */
110 unsigned short hash
; /* hash value (index in ht[]) */
111 short allot
; /* credit for this slot */
113 unsigned int backlog
;
114 struct red_vars vars
;
117 struct sfq_sched_data
{
118 /* frequently used fields */
119 int limit
; /* limit of total number of packets in this qdisc */
120 unsigned int divisor
; /* number of slots in hash table */
122 u8 maxdepth
; /* limit of packets per flow */
125 u8 cur_depth
; /* depth of longest slot */
127 unsigned short scaled_quantum
; /* SFQ_ALLOT_SIZE(quantum) */
128 struct tcf_proto __rcu
*filter_list
;
129 sfq_index
*ht
; /* Hash table ('divisor' slots) */
130 struct sfq_slot
*slots
; /* Flows table ('maxflows' entries) */
132 struct red_parms
*red_parms
;
133 struct tc_sfqred_stats stats
;
134 struct sfq_slot
*tail
; /* current slot in round */
136 struct sfq_head dep
[SFQ_MAX_DEPTH
+ 1];
137 /* Linked lists of slots, indexed by depth
138 * dep[0] : list of unused flows
139 * dep[1] : list of flows with 1 packet
140 * dep[X] : list of flows with X packets
143 unsigned int maxflows
; /* number of flows in flows array */
145 unsigned int quantum
; /* Allotment per round: MUST BE >= MTU */
146 struct timer_list perturb_timer
;
150 * sfq_head are either in a sfq_slot or in dep[] array
152 static inline struct sfq_head
*sfq_dep_head(struct sfq_sched_data
*q
, sfq_index val
)
154 if (val
< SFQ_MAX_FLOWS
)
155 return &q
->slots
[val
].dep
;
156 return &q
->dep
[val
- SFQ_MAX_FLOWS
];
160 * In order to be able to quickly rehash our queue when timer changes
161 * q->perturbation, we store flow_keys in skb->cb[]
164 struct flow_keys keys
;
167 static inline struct sfq_skb_cb
*sfq_skb_cb(const struct sk_buff
*skb
)
169 qdisc_cb_private_validate(skb
, sizeof(struct sfq_skb_cb
));
170 return (struct sfq_skb_cb
*)qdisc_skb_cb(skb
)->data
;
173 static unsigned int sfq_hash(const struct sfq_sched_data
*q
,
174 const struct sk_buff
*skb
)
176 const struct flow_keys
*keys
= &sfq_skb_cb(skb
)->keys
;
179 hash
= jhash_3words((__force u32
)keys
->dst
,
180 (__force u32
)keys
->src
^ keys
->ip_proto
,
181 (__force u32
)keys
->ports
, q
->perturbation
);
182 return hash
& (q
->divisor
- 1);
185 static unsigned int sfq_classify(struct sk_buff
*skb
, struct Qdisc
*sch
,
188 struct sfq_sched_data
*q
= qdisc_priv(sch
);
189 struct tcf_result res
;
190 struct tcf_proto
*fl
;
193 if (TC_H_MAJ(skb
->priority
) == sch
->handle
&&
194 TC_H_MIN(skb
->priority
) > 0 &&
195 TC_H_MIN(skb
->priority
) <= q
->divisor
)
196 return TC_H_MIN(skb
->priority
);
198 fl
= rcu_dereference_bh(q
->filter_list
);
200 skb_flow_dissect(skb
, &sfq_skb_cb(skb
)->keys
);
201 return sfq_hash(q
, skb
) + 1;
204 *qerr
= NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
205 result
= tc_classify(skb
, fl
, &res
);
207 #ifdef CONFIG_NET_CLS_ACT
211 *qerr
= NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
216 if (TC_H_MIN(res
.classid
) <= q
->divisor
)
217 return TC_H_MIN(res
.classid
);
223 * x : slot number [0 .. SFQ_MAX_FLOWS - 1]
225 static inline void sfq_link(struct sfq_sched_data
*q
, sfq_index x
)
228 struct sfq_slot
*slot
= &q
->slots
[x
];
229 int qlen
= slot
->qlen
;
231 p
= qlen
+ SFQ_MAX_FLOWS
;
232 n
= q
->dep
[qlen
].next
;
237 q
->dep
[qlen
].next
= x
; /* sfq_dep_head(q, p)->next = x */
238 sfq_dep_head(q
, n
)->prev
= x
;
241 #define sfq_unlink(q, x, n, p) \
243 n = q->slots[x].dep.next; \
244 p = q->slots[x].dep.prev; \
245 sfq_dep_head(q, p)->next = n; \
246 sfq_dep_head(q, n)->prev = p; \
250 static inline void sfq_dec(struct sfq_sched_data
*q
, sfq_index x
)
255 sfq_unlink(q
, x
, n
, p
);
257 d
= q
->slots
[x
].qlen
--;
258 if (n
== p
&& q
->cur_depth
== d
)
263 static inline void sfq_inc(struct sfq_sched_data
*q
, sfq_index x
)
268 sfq_unlink(q
, x
, n
, p
);
270 d
= ++q
->slots
[x
].qlen
;
271 if (q
->cur_depth
< d
)
276 /* helper functions : might be changed when/if skb use a standard list_head */
278 /* remove one skb from tail of slot queue */
279 static inline struct sk_buff
*slot_dequeue_tail(struct sfq_slot
*slot
)
281 struct sk_buff
*skb
= slot
->skblist_prev
;
283 slot
->skblist_prev
= skb
->prev
;
284 skb
->prev
->next
= (struct sk_buff
*)slot
;
285 skb
->next
= skb
->prev
= NULL
;
289 /* remove one skb from head of slot queue */
290 static inline struct sk_buff
*slot_dequeue_head(struct sfq_slot
*slot
)
292 struct sk_buff
*skb
= slot
->skblist_next
;
294 slot
->skblist_next
= skb
->next
;
295 skb
->next
->prev
= (struct sk_buff
*)slot
;
296 skb
->next
= skb
->prev
= NULL
;
300 static inline void slot_queue_init(struct sfq_slot
*slot
)
302 memset(slot
, 0, sizeof(*slot
));
303 slot
->skblist_prev
= slot
->skblist_next
= (struct sk_buff
*)slot
;
306 /* add skb to slot queue (tail add) */
307 static inline void slot_queue_add(struct sfq_slot
*slot
, struct sk_buff
*skb
)
309 skb
->prev
= slot
->skblist_prev
;
310 skb
->next
= (struct sk_buff
*)slot
;
311 slot
->skblist_prev
->next
= skb
;
312 slot
->skblist_prev
= skb
;
315 static unsigned int sfq_drop(struct Qdisc
*sch
)
317 struct sfq_sched_data
*q
= qdisc_priv(sch
);
318 sfq_index x
, d
= q
->cur_depth
;
321 struct sfq_slot
*slot
;
323 /* Queue is full! Find the longest slot and drop tail packet from it */
328 skb
= q
->headdrop
? slot_dequeue_head(slot
) : slot_dequeue_tail(slot
);
329 len
= qdisc_pkt_len(skb
);
330 slot
->backlog
-= len
;
335 sch
->qstats
.backlog
-= len
;
340 /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
343 q
->tail
->next
= slot
->next
;
344 q
->ht
[slot
->hash
] = SFQ_EMPTY_SLOT
;
351 /* Is ECN parameter configured */
352 static int sfq_prob_mark(const struct sfq_sched_data
*q
)
354 return q
->flags
& TC_RED_ECN
;
357 /* Should packets over max threshold just be marked */
358 static int sfq_hard_mark(const struct sfq_sched_data
*q
)
360 return (q
->flags
& (TC_RED_ECN
| TC_RED_HARDDROP
)) == TC_RED_ECN
;
363 static int sfq_headdrop(const struct sfq_sched_data
*q
)
369 sfq_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
)
371 struct sfq_sched_data
*q
= qdisc_priv(sch
);
374 struct sfq_slot
*slot
;
375 int uninitialized_var(ret
);
376 struct sk_buff
*head
;
379 hash
= sfq_classify(skb
, sch
, &ret
);
381 if (ret
& __NET_XMIT_BYPASS
)
390 if (x
== SFQ_EMPTY_SLOT
) {
391 x
= q
->dep
[0].next
; /* get a free slot */
392 if (x
>= SFQ_MAX_FLOWS
)
393 return qdisc_drop(skb
, sch
);
397 slot
->backlog
= 0; /* should already be 0 anyway... */
398 red_set_vars(&slot
->vars
);
402 slot
->vars
.qavg
= red_calc_qavg_no_idle_time(q
->red_parms
,
405 switch (red_action(q
->red_parms
,
412 sch
->qstats
.overlimits
++;
413 if (sfq_prob_mark(q
)) {
414 /* We know we have at least one packet in queue */
415 if (sfq_headdrop(q
) &&
416 INET_ECN_set_ce(slot
->skblist_next
)) {
417 q
->stats
.prob_mark_head
++;
420 if (INET_ECN_set_ce(skb
)) {
421 q
->stats
.prob_mark
++;
425 q
->stats
.prob_drop
++;
426 goto congestion_drop
;
429 sch
->qstats
.overlimits
++;
430 if (sfq_hard_mark(q
)) {
431 /* We know we have at least one packet in queue */
432 if (sfq_headdrop(q
) &&
433 INET_ECN_set_ce(slot
->skblist_next
)) {
434 q
->stats
.forced_mark_head
++;
437 if (INET_ECN_set_ce(skb
)) {
438 q
->stats
.forced_mark
++;
442 q
->stats
.forced_drop
++;
443 goto congestion_drop
;
447 if (slot
->qlen
>= q
->maxdepth
) {
449 if (!sfq_headdrop(q
))
450 return qdisc_drop(skb
, sch
);
452 /* We know we have at least one packet in queue */
453 head
= slot_dequeue_head(slot
);
454 delta
= qdisc_pkt_len(head
) - qdisc_pkt_len(skb
);
455 sch
->qstats
.backlog
-= delta
;
456 slot
->backlog
-= delta
;
457 qdisc_drop(head
, sch
);
459 slot_queue_add(slot
, skb
);
464 sch
->qstats
.backlog
+= qdisc_pkt_len(skb
);
465 slot
->backlog
+= qdisc_pkt_len(skb
);
466 slot_queue_add(slot
, skb
);
468 if (slot
->qlen
== 1) { /* The flow is new */
469 if (q
->tail
== NULL
) { /* It is the first flow */
472 slot
->next
= q
->tail
->next
;
475 /* We put this flow at the end of our flow list.
476 * This might sound unfair for a new flow to wait after old ones,
477 * but we could endup servicing new flows only, and freeze old ones.
480 /* We could use a bigger initial quantum for new flows */
481 slot
->allot
= q
->scaled_quantum
;
483 if (++sch
->q
.qlen
<= q
->limit
)
484 return NET_XMIT_SUCCESS
;
488 /* Return Congestion Notification only if we dropped a packet
491 if (qlen
!= slot
->qlen
)
494 /* As we dropped a packet, better let upper stack know this */
495 qdisc_tree_decrease_qlen(sch
, 1);
496 return NET_XMIT_SUCCESS
;
499 static struct sk_buff
*
500 sfq_dequeue(struct Qdisc
*sch
)
502 struct sfq_sched_data
*q
= qdisc_priv(sch
);
505 struct sfq_slot
*slot
;
507 /* No active slots */
514 if (slot
->allot
<= 0) {
516 slot
->allot
+= q
->scaled_quantum
;
519 skb
= slot_dequeue_head(slot
);
521 qdisc_bstats_update(sch
, skb
);
523 sch
->qstats
.backlog
-= qdisc_pkt_len(skb
);
524 slot
->backlog
-= qdisc_pkt_len(skb
);
525 /* Is the slot empty? */
526 if (slot
->qlen
== 0) {
527 q
->ht
[slot
->hash
] = SFQ_EMPTY_SLOT
;
530 q
->tail
= NULL
; /* no more active slots */
533 q
->tail
->next
= next_a
;
535 slot
->allot
-= SFQ_ALLOT_SIZE(qdisc_pkt_len(skb
));
541 sfq_reset(struct Qdisc
*sch
)
545 while ((skb
= sfq_dequeue(sch
)) != NULL
)
550 * When q->perturbation is changed, we rehash all queued skbs
551 * to avoid OOO (Out Of Order) effects.
552 * We dont use sfq_dequeue()/sfq_enqueue() because we dont want to change
555 static void sfq_rehash(struct Qdisc
*sch
)
557 struct sfq_sched_data
*q
= qdisc_priv(sch
);
560 struct sfq_slot
*slot
;
561 struct sk_buff_head list
;
564 __skb_queue_head_init(&list
);
566 for (i
= 0; i
< q
->maxflows
; i
++) {
571 skb
= slot_dequeue_head(slot
);
573 __skb_queue_tail(&list
, skb
);
576 red_set_vars(&slot
->vars
);
577 q
->ht
[slot
->hash
] = SFQ_EMPTY_SLOT
;
581 while ((skb
= __skb_dequeue(&list
)) != NULL
) {
582 unsigned int hash
= sfq_hash(q
, skb
);
583 sfq_index x
= q
->ht
[hash
];
586 if (x
== SFQ_EMPTY_SLOT
) {
587 x
= q
->dep
[0].next
; /* get a free slot */
588 if (x
>= SFQ_MAX_FLOWS
) {
589 drop
: sch
->qstats
.backlog
-= qdisc_pkt_len(skb
);
598 if (slot
->qlen
>= q
->maxdepth
)
600 slot_queue_add(slot
, skb
);
602 slot
->vars
.qavg
= red_calc_qavg(q
->red_parms
,
605 slot
->backlog
+= qdisc_pkt_len(skb
);
607 if (slot
->qlen
== 1) { /* The flow is new */
608 if (q
->tail
== NULL
) { /* It is the first flow */
611 slot
->next
= q
->tail
->next
;
615 slot
->allot
= q
->scaled_quantum
;
618 sch
->q
.qlen
-= dropped
;
619 qdisc_tree_decrease_qlen(sch
, dropped
);
622 static void sfq_perturbation(unsigned long arg
)
624 struct Qdisc
*sch
= (struct Qdisc
*)arg
;
625 struct sfq_sched_data
*q
= qdisc_priv(sch
);
626 spinlock_t
*root_lock
= qdisc_lock(qdisc_root_sleeping(sch
));
628 spin_lock(root_lock
);
629 q
->perturbation
= prandom_u32();
630 if (!q
->filter_list
&& q
->tail
)
632 spin_unlock(root_lock
);
634 if (q
->perturb_period
)
635 mod_timer(&q
->perturb_timer
, jiffies
+ q
->perturb_period
);
638 static int sfq_change(struct Qdisc
*sch
, struct nlattr
*opt
)
640 struct sfq_sched_data
*q
= qdisc_priv(sch
);
641 struct tc_sfq_qopt
*ctl
= nla_data(opt
);
642 struct tc_sfq_qopt_v1
*ctl_v1
= NULL
;
644 struct red_parms
*p
= NULL
;
646 if (opt
->nla_len
< nla_attr_size(sizeof(*ctl
)))
648 if (opt
->nla_len
>= nla_attr_size(sizeof(*ctl_v1
)))
649 ctl_v1
= nla_data(opt
);
651 (!is_power_of_2(ctl
->divisor
) || ctl
->divisor
> 65536))
653 if (ctl_v1
&& ctl_v1
->qth_min
) {
654 p
= kmalloc(sizeof(*p
), GFP_KERNEL
);
660 q
->quantum
= ctl
->quantum
;
661 q
->scaled_quantum
= SFQ_ALLOT_SIZE(q
->quantum
);
663 q
->perturb_period
= ctl
->perturb_period
* HZ
;
665 q
->maxflows
= min_t(u32
, ctl
->flows
, SFQ_MAX_FLOWS
);
667 q
->divisor
= ctl
->divisor
;
668 q
->maxflows
= min_t(u32
, q
->maxflows
, q
->divisor
);
672 q
->maxdepth
= min_t(u32
, ctl_v1
->depth
, SFQ_MAX_DEPTH
);
674 swap(q
->red_parms
, p
);
675 red_set_parms(q
->red_parms
,
676 ctl_v1
->qth_min
, ctl_v1
->qth_max
,
678 ctl_v1
->Plog
, ctl_v1
->Scell_log
,
682 q
->flags
= ctl_v1
->flags
;
683 q
->headdrop
= ctl_v1
->headdrop
;
686 q
->limit
= min_t(u32
, ctl
->limit
, q
->maxdepth
* q
->maxflows
);
687 q
->maxflows
= min_t(u32
, q
->maxflows
, q
->limit
);
691 while (sch
->q
.qlen
> q
->limit
)
693 qdisc_tree_decrease_qlen(sch
, qlen
- sch
->q
.qlen
);
695 del_timer(&q
->perturb_timer
);
696 if (q
->perturb_period
) {
697 mod_timer(&q
->perturb_timer
, jiffies
+ q
->perturb_period
);
698 q
->perturbation
= prandom_u32();
700 sch_tree_unlock(sch
);
705 static void *sfq_alloc(size_t sz
)
707 void *ptr
= kmalloc(sz
, GFP_KERNEL
| __GFP_NOWARN
);
714 static void sfq_free(void *addr
)
719 static void sfq_destroy(struct Qdisc
*sch
)
721 struct sfq_sched_data
*q
= qdisc_priv(sch
);
723 tcf_destroy_chain(&q
->filter_list
);
724 q
->perturb_period
= 0;
725 del_timer_sync(&q
->perturb_timer
);
731 static int sfq_init(struct Qdisc
*sch
, struct nlattr
*opt
)
733 struct sfq_sched_data
*q
= qdisc_priv(sch
);
736 q
->perturb_timer
.function
= sfq_perturbation
;
737 q
->perturb_timer
.data
= (unsigned long)sch
;
738 init_timer_deferrable(&q
->perturb_timer
);
740 for (i
= 0; i
< SFQ_MAX_DEPTH
+ 1; i
++) {
741 q
->dep
[i
].next
= i
+ SFQ_MAX_FLOWS
;
742 q
->dep
[i
].prev
= i
+ SFQ_MAX_FLOWS
;
745 q
->limit
= SFQ_MAX_DEPTH
;
746 q
->maxdepth
= SFQ_MAX_DEPTH
;
749 q
->divisor
= SFQ_DEFAULT_HASH_DIVISOR
;
750 q
->maxflows
= SFQ_DEFAULT_FLOWS
;
751 q
->quantum
= psched_mtu(qdisc_dev(sch
));
752 q
->scaled_quantum
= SFQ_ALLOT_SIZE(q
->quantum
);
753 q
->perturb_period
= 0;
754 q
->perturbation
= prandom_u32();
757 int err
= sfq_change(sch
, opt
);
762 q
->ht
= sfq_alloc(sizeof(q
->ht
[0]) * q
->divisor
);
763 q
->slots
= sfq_alloc(sizeof(q
->slots
[0]) * q
->maxflows
);
764 if (!q
->ht
|| !q
->slots
) {
768 for (i
= 0; i
< q
->divisor
; i
++)
769 q
->ht
[i
] = SFQ_EMPTY_SLOT
;
771 for (i
= 0; i
< q
->maxflows
; i
++) {
772 slot_queue_init(&q
->slots
[i
]);
776 sch
->flags
|= TCQ_F_CAN_BYPASS
;
778 sch
->flags
&= ~TCQ_F_CAN_BYPASS
;
782 static int sfq_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
784 struct sfq_sched_data
*q
= qdisc_priv(sch
);
785 unsigned char *b
= skb_tail_pointer(skb
);
786 struct tc_sfq_qopt_v1 opt
;
787 struct red_parms
*p
= q
->red_parms
;
789 memset(&opt
, 0, sizeof(opt
));
790 opt
.v0
.quantum
= q
->quantum
;
791 opt
.v0
.perturb_period
= q
->perturb_period
/ HZ
;
792 opt
.v0
.limit
= q
->limit
;
793 opt
.v0
.divisor
= q
->divisor
;
794 opt
.v0
.flows
= q
->maxflows
;
795 opt
.depth
= q
->maxdepth
;
796 opt
.headdrop
= q
->headdrop
;
799 opt
.qth_min
= p
->qth_min
>> p
->Wlog
;
800 opt
.qth_max
= p
->qth_max
>> p
->Wlog
;
803 opt
.Scell_log
= p
->Scell_log
;
804 opt
.max_P
= p
->max_P
;
806 memcpy(&opt
.stats
, &q
->stats
, sizeof(opt
.stats
));
807 opt
.flags
= q
->flags
;
809 if (nla_put(skb
, TCA_OPTIONS
, sizeof(opt
), &opt
))
810 goto nla_put_failure
;
819 static struct Qdisc
*sfq_leaf(struct Qdisc
*sch
, unsigned long arg
)
824 static unsigned long sfq_get(struct Qdisc
*sch
, u32 classid
)
829 static unsigned long sfq_bind(struct Qdisc
*sch
, unsigned long parent
,
832 /* we cannot bypass queue discipline anymore */
833 sch
->flags
&= ~TCQ_F_CAN_BYPASS
;
837 static void sfq_put(struct Qdisc
*q
, unsigned long cl
)
841 static struct tcf_proto __rcu
**sfq_find_tcf(struct Qdisc
*sch
,
844 struct sfq_sched_data
*q
= qdisc_priv(sch
);
848 return &q
->filter_list
;
851 static int sfq_dump_class(struct Qdisc
*sch
, unsigned long cl
,
852 struct sk_buff
*skb
, struct tcmsg
*tcm
)
854 tcm
->tcm_handle
|= TC_H_MIN(cl
);
858 static int sfq_dump_class_stats(struct Qdisc
*sch
, unsigned long cl
,
861 struct sfq_sched_data
*q
= qdisc_priv(sch
);
862 sfq_index idx
= q
->ht
[cl
- 1];
863 struct gnet_stats_queue qs
= { 0 };
864 struct tc_sfq_xstats xstats
= { 0 };
866 if (idx
!= SFQ_EMPTY_SLOT
) {
867 const struct sfq_slot
*slot
= &q
->slots
[idx
];
869 xstats
.allot
= slot
->allot
<< SFQ_ALLOT_SHIFT
;
870 qs
.qlen
= slot
->qlen
;
871 qs
.backlog
= slot
->backlog
;
873 if (gnet_stats_copy_queue(d
, &qs
) < 0)
875 return gnet_stats_copy_app(d
, &xstats
, sizeof(xstats
));
878 static void sfq_walk(struct Qdisc
*sch
, struct qdisc_walker
*arg
)
880 struct sfq_sched_data
*q
= qdisc_priv(sch
);
886 for (i
= 0; i
< q
->divisor
; i
++) {
887 if (q
->ht
[i
] == SFQ_EMPTY_SLOT
||
888 arg
->count
< arg
->skip
) {
892 if (arg
->fn(sch
, i
+ 1, arg
) < 0) {
900 static const struct Qdisc_class_ops sfq_class_ops
= {
904 .tcf_chain
= sfq_find_tcf
,
905 .bind_tcf
= sfq_bind
,
906 .unbind_tcf
= sfq_put
,
907 .dump
= sfq_dump_class
,
908 .dump_stats
= sfq_dump_class_stats
,
912 static struct Qdisc_ops sfq_qdisc_ops __read_mostly
= {
913 .cl_ops
= &sfq_class_ops
,
915 .priv_size
= sizeof(struct sfq_sched_data
),
916 .enqueue
= sfq_enqueue
,
917 .dequeue
= sfq_dequeue
,
918 .peek
= qdisc_peek_dequeued
,
922 .destroy
= sfq_destroy
,
925 .owner
= THIS_MODULE
,
928 static int __init
sfq_module_init(void)
930 return register_qdisc(&sfq_qdisc_ops
);
932 static void __exit
sfq_module_exit(void)
934 unregister_qdisc(&sfq_qdisc_ops
);
936 module_init(sfq_module_init
)
937 module_exit(sfq_module_exit
)
938 MODULE_LICENSE("GPL");