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/siphash.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/pkt_cls.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 */
124 siphash_key_t perturbation
;
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 struct tcf_block
*block
;
130 sfq_index
*ht
; /* Hash table ('divisor' slots) */
131 struct sfq_slot
*slots
; /* Flows table ('maxflows' entries) */
133 struct red_parms
*red_parms
;
134 struct tc_sfqred_stats stats
;
135 struct sfq_slot
*tail
; /* current slot in round */
137 struct sfq_head dep
[SFQ_MAX_DEPTH
+ 1];
138 /* Linked lists of slots, indexed by depth
139 * dep[0] : list of unused flows
140 * dep[1] : list of flows with 1 packet
141 * dep[X] : list of flows with X packets
144 unsigned int maxflows
; /* number of flows in flows array */
146 unsigned int quantum
; /* Allotment per round: MUST BE >= MTU */
147 struct timer_list perturb_timer
;
151 * sfq_head are either in a sfq_slot or in dep[] array
153 static inline struct sfq_head
*sfq_dep_head(struct sfq_sched_data
*q
, sfq_index val
)
155 if (val
< SFQ_MAX_FLOWS
)
156 return &q
->slots
[val
].dep
;
157 return &q
->dep
[val
- SFQ_MAX_FLOWS
];
160 static unsigned int sfq_hash(const struct sfq_sched_data
*q
,
161 const struct sk_buff
*skb
)
163 return skb_get_hash_perturb(skb
, &q
->perturbation
) & (q
->divisor
- 1);
166 static unsigned int sfq_classify(struct sk_buff
*skb
, struct Qdisc
*sch
,
169 struct sfq_sched_data
*q
= qdisc_priv(sch
);
170 struct tcf_result res
;
171 struct tcf_proto
*fl
;
174 if (TC_H_MAJ(skb
->priority
) == sch
->handle
&&
175 TC_H_MIN(skb
->priority
) > 0 &&
176 TC_H_MIN(skb
->priority
) <= q
->divisor
)
177 return TC_H_MIN(skb
->priority
);
179 fl
= rcu_dereference_bh(q
->filter_list
);
181 return sfq_hash(q
, skb
) + 1;
183 *qerr
= NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
184 result
= tcf_classify(skb
, fl
, &res
, false);
186 #ifdef CONFIG_NET_CLS_ACT
191 *qerr
= NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
196 if (TC_H_MIN(res
.classid
) <= q
->divisor
)
197 return TC_H_MIN(res
.classid
);
203 * x : slot number [0 .. SFQ_MAX_FLOWS - 1]
205 static inline void sfq_link(struct sfq_sched_data
*q
, sfq_index x
)
208 struct sfq_slot
*slot
= &q
->slots
[x
];
209 int qlen
= slot
->qlen
;
211 p
= qlen
+ SFQ_MAX_FLOWS
;
212 n
= q
->dep
[qlen
].next
;
217 q
->dep
[qlen
].next
= x
; /* sfq_dep_head(q, p)->next = x */
218 sfq_dep_head(q
, n
)->prev
= x
;
221 #define sfq_unlink(q, x, n, p) \
223 n = q->slots[x].dep.next; \
224 p = q->slots[x].dep.prev; \
225 sfq_dep_head(q, p)->next = n; \
226 sfq_dep_head(q, n)->prev = p; \
230 static inline void sfq_dec(struct sfq_sched_data
*q
, sfq_index x
)
235 sfq_unlink(q
, x
, n
, p
);
237 d
= q
->slots
[x
].qlen
--;
238 if (n
== p
&& q
->cur_depth
== d
)
243 static inline void sfq_inc(struct sfq_sched_data
*q
, sfq_index x
)
248 sfq_unlink(q
, x
, n
, p
);
250 d
= ++q
->slots
[x
].qlen
;
251 if (q
->cur_depth
< d
)
256 /* helper functions : might be changed when/if skb use a standard list_head */
258 /* remove one skb from tail of slot queue */
259 static inline struct sk_buff
*slot_dequeue_tail(struct sfq_slot
*slot
)
261 struct sk_buff
*skb
= slot
->skblist_prev
;
263 slot
->skblist_prev
= skb
->prev
;
264 skb
->prev
->next
= (struct sk_buff
*)slot
;
265 skb
->next
= skb
->prev
= NULL
;
269 /* remove one skb from head of slot queue */
270 static inline struct sk_buff
*slot_dequeue_head(struct sfq_slot
*slot
)
272 struct sk_buff
*skb
= slot
->skblist_next
;
274 slot
->skblist_next
= skb
->next
;
275 skb
->next
->prev
= (struct sk_buff
*)slot
;
276 skb
->next
= skb
->prev
= NULL
;
280 static inline void slot_queue_init(struct sfq_slot
*slot
)
282 memset(slot
, 0, sizeof(*slot
));
283 slot
->skblist_prev
= slot
->skblist_next
= (struct sk_buff
*)slot
;
286 /* add skb to slot queue (tail add) */
287 static inline void slot_queue_add(struct sfq_slot
*slot
, struct sk_buff
*skb
)
289 skb
->prev
= slot
->skblist_prev
;
290 skb
->next
= (struct sk_buff
*)slot
;
291 slot
->skblist_prev
->next
= skb
;
292 slot
->skblist_prev
= skb
;
295 static unsigned int sfq_drop(struct Qdisc
*sch
, struct sk_buff
**to_free
)
297 struct sfq_sched_data
*q
= qdisc_priv(sch
);
298 sfq_index x
, d
= q
->cur_depth
;
301 struct sfq_slot
*slot
;
303 /* Queue is full! Find the longest slot and drop tail packet from it */
308 skb
= q
->headdrop
? slot_dequeue_head(slot
) : slot_dequeue_tail(slot
);
309 len
= qdisc_pkt_len(skb
);
310 slot
->backlog
-= len
;
313 qdisc_qstats_backlog_dec(sch
, skb
);
314 qdisc_drop(skb
, sch
, to_free
);
319 /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
322 q
->tail
->next
= slot
->next
;
323 q
->ht
[slot
->hash
] = SFQ_EMPTY_SLOT
;
330 /* Is ECN parameter configured */
331 static int sfq_prob_mark(const struct sfq_sched_data
*q
)
333 return q
->flags
& TC_RED_ECN
;
336 /* Should packets over max threshold just be marked */
337 static int sfq_hard_mark(const struct sfq_sched_data
*q
)
339 return (q
->flags
& (TC_RED_ECN
| TC_RED_HARDDROP
)) == TC_RED_ECN
;
342 static int sfq_headdrop(const struct sfq_sched_data
*q
)
348 sfq_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
, struct sk_buff
**to_free
)
350 struct sfq_sched_data
*q
= qdisc_priv(sch
);
351 unsigned int hash
, dropped
;
353 struct sfq_slot
*slot
;
354 int uninitialized_var(ret
);
355 struct sk_buff
*head
;
358 hash
= sfq_classify(skb
, sch
, &ret
);
360 if (ret
& __NET_XMIT_BYPASS
)
361 qdisc_qstats_drop(sch
);
362 __qdisc_drop(skb
, to_free
);
369 if (x
== SFQ_EMPTY_SLOT
) {
370 x
= q
->dep
[0].next
; /* get a free slot */
371 if (x
>= SFQ_MAX_FLOWS
)
372 return qdisc_drop(skb
, sch
, to_free
);
376 slot
->backlog
= 0; /* should already be 0 anyway... */
377 red_set_vars(&slot
->vars
);
381 slot
->vars
.qavg
= red_calc_qavg_no_idle_time(q
->red_parms
,
384 switch (red_action(q
->red_parms
,
391 qdisc_qstats_overlimit(sch
);
392 if (sfq_prob_mark(q
)) {
393 /* We know we have at least one packet in queue */
394 if (sfq_headdrop(q
) &&
395 INET_ECN_set_ce(slot
->skblist_next
)) {
396 q
->stats
.prob_mark_head
++;
399 if (INET_ECN_set_ce(skb
)) {
400 q
->stats
.prob_mark
++;
404 q
->stats
.prob_drop
++;
405 goto congestion_drop
;
408 qdisc_qstats_overlimit(sch
);
409 if (sfq_hard_mark(q
)) {
410 /* We know we have at least one packet in queue */
411 if (sfq_headdrop(q
) &&
412 INET_ECN_set_ce(slot
->skblist_next
)) {
413 q
->stats
.forced_mark_head
++;
416 if (INET_ECN_set_ce(skb
)) {
417 q
->stats
.forced_mark
++;
421 q
->stats
.forced_drop
++;
422 goto congestion_drop
;
426 if (slot
->qlen
>= q
->maxdepth
) {
428 if (!sfq_headdrop(q
))
429 return qdisc_drop(skb
, sch
, to_free
);
431 /* We know we have at least one packet in queue */
432 head
= slot_dequeue_head(slot
);
433 delta
= qdisc_pkt_len(head
) - qdisc_pkt_len(skb
);
434 sch
->qstats
.backlog
-= delta
;
435 slot
->backlog
-= delta
;
436 qdisc_drop(head
, sch
, to_free
);
438 slot_queue_add(slot
, skb
);
439 qdisc_tree_reduce_backlog(sch
, 0, delta
);
444 qdisc_qstats_backlog_inc(sch
, skb
);
445 slot
->backlog
+= qdisc_pkt_len(skb
);
446 slot_queue_add(slot
, skb
);
448 if (slot
->qlen
== 1) { /* The flow is new */
449 if (q
->tail
== NULL
) { /* It is the first flow */
452 slot
->next
= q
->tail
->next
;
455 /* We put this flow at the end of our flow list.
456 * This might sound unfair for a new flow to wait after old ones,
457 * but we could endup servicing new flows only, and freeze old ones.
460 /* We could use a bigger initial quantum for new flows */
461 slot
->allot
= q
->scaled_quantum
;
463 if (++sch
->q
.qlen
<= q
->limit
)
464 return NET_XMIT_SUCCESS
;
467 dropped
= sfq_drop(sch
, to_free
);
468 /* Return Congestion Notification only if we dropped a packet
471 if (qlen
!= slot
->qlen
) {
472 qdisc_tree_reduce_backlog(sch
, 0, dropped
- qdisc_pkt_len(skb
));
476 /* As we dropped a packet, better let upper stack know this */
477 qdisc_tree_reduce_backlog(sch
, 1, dropped
);
478 return NET_XMIT_SUCCESS
;
481 static struct sk_buff
*
482 sfq_dequeue(struct Qdisc
*sch
)
484 struct sfq_sched_data
*q
= qdisc_priv(sch
);
487 struct sfq_slot
*slot
;
489 /* No active slots */
496 if (slot
->allot
<= 0) {
498 slot
->allot
+= q
->scaled_quantum
;
501 skb
= slot_dequeue_head(slot
);
503 qdisc_bstats_update(sch
, skb
);
505 qdisc_qstats_backlog_dec(sch
, skb
);
506 slot
->backlog
-= qdisc_pkt_len(skb
);
507 /* Is the slot empty? */
508 if (slot
->qlen
== 0) {
509 q
->ht
[slot
->hash
] = SFQ_EMPTY_SLOT
;
512 q
->tail
= NULL
; /* no more active slots */
515 q
->tail
->next
= next_a
;
517 slot
->allot
-= SFQ_ALLOT_SIZE(qdisc_pkt_len(skb
));
523 sfq_reset(struct Qdisc
*sch
)
527 while ((skb
= sfq_dequeue(sch
)) != NULL
)
528 rtnl_kfree_skbs(skb
, skb
);
532 * When q->perturbation is changed, we rehash all queued skbs
533 * to avoid OOO (Out Of Order) effects.
534 * We dont use sfq_dequeue()/sfq_enqueue() because we dont want to change
537 static void sfq_rehash(struct Qdisc
*sch
)
539 struct sfq_sched_data
*q
= qdisc_priv(sch
);
542 struct sfq_slot
*slot
;
543 struct sk_buff_head list
;
545 unsigned int drop_len
= 0;
547 __skb_queue_head_init(&list
);
549 for (i
= 0; i
< q
->maxflows
; i
++) {
554 skb
= slot_dequeue_head(slot
);
556 __skb_queue_tail(&list
, skb
);
559 red_set_vars(&slot
->vars
);
560 q
->ht
[slot
->hash
] = SFQ_EMPTY_SLOT
;
564 while ((skb
= __skb_dequeue(&list
)) != NULL
) {
565 unsigned int hash
= sfq_hash(q
, skb
);
566 sfq_index x
= q
->ht
[hash
];
569 if (x
== SFQ_EMPTY_SLOT
) {
570 x
= q
->dep
[0].next
; /* get a free slot */
571 if (x
>= SFQ_MAX_FLOWS
) {
573 qdisc_qstats_backlog_dec(sch
, skb
);
574 drop_len
+= qdisc_pkt_len(skb
);
583 if (slot
->qlen
>= q
->maxdepth
)
585 slot_queue_add(slot
, skb
);
587 slot
->vars
.qavg
= red_calc_qavg(q
->red_parms
,
590 slot
->backlog
+= qdisc_pkt_len(skb
);
592 if (slot
->qlen
== 1) { /* The flow is new */
593 if (q
->tail
== NULL
) { /* It is the first flow */
596 slot
->next
= q
->tail
->next
;
600 slot
->allot
= q
->scaled_quantum
;
603 sch
->q
.qlen
-= dropped
;
604 qdisc_tree_reduce_backlog(sch
, dropped
, drop_len
);
607 static void sfq_perturbation(unsigned long arg
)
609 struct Qdisc
*sch
= (struct Qdisc
*)arg
;
610 struct sfq_sched_data
*q
= qdisc_priv(sch
);
611 spinlock_t
*root_lock
= qdisc_lock(qdisc_root_sleeping(sch
));
614 get_random_bytes(&nkey
, sizeof(nkey
));
615 spin_lock(root_lock
);
616 q
->perturbation
= nkey
;
617 if (!q
->filter_list
&& q
->tail
)
619 spin_unlock(root_lock
);
621 if (q
->perturb_period
)
622 mod_timer(&q
->perturb_timer
, jiffies
+ q
->perturb_period
);
625 static int sfq_change(struct Qdisc
*sch
, struct nlattr
*opt
)
627 struct sfq_sched_data
*q
= qdisc_priv(sch
);
628 struct tc_sfq_qopt
*ctl
= nla_data(opt
);
629 struct tc_sfq_qopt_v1
*ctl_v1
= NULL
;
630 unsigned int qlen
, dropped
= 0;
631 struct red_parms
*p
= NULL
;
632 struct sk_buff
*to_free
= NULL
;
633 struct sk_buff
*tail
= NULL
;
635 if (opt
->nla_len
< nla_attr_size(sizeof(*ctl
)))
637 if (opt
->nla_len
>= nla_attr_size(sizeof(*ctl_v1
)))
638 ctl_v1
= nla_data(opt
);
640 (!is_power_of_2(ctl
->divisor
) || ctl
->divisor
> 65536))
642 if (ctl_v1
&& !red_check_params(ctl_v1
->qth_min
, ctl_v1
->qth_max
,
645 if (ctl_v1
&& ctl_v1
->qth_min
) {
646 p
= kmalloc(sizeof(*p
), GFP_KERNEL
);
652 q
->quantum
= ctl
->quantum
;
653 q
->scaled_quantum
= SFQ_ALLOT_SIZE(q
->quantum
);
655 q
->perturb_period
= ctl
->perturb_period
* HZ
;
657 q
->maxflows
= min_t(u32
, ctl
->flows
, SFQ_MAX_FLOWS
);
659 q
->divisor
= ctl
->divisor
;
660 q
->maxflows
= min_t(u32
, q
->maxflows
, q
->divisor
);
664 q
->maxdepth
= min_t(u32
, ctl_v1
->depth
, SFQ_MAX_DEPTH
);
666 swap(q
->red_parms
, p
);
667 red_set_parms(q
->red_parms
,
668 ctl_v1
->qth_min
, ctl_v1
->qth_max
,
670 ctl_v1
->Plog
, ctl_v1
->Scell_log
,
674 q
->flags
= ctl_v1
->flags
;
675 q
->headdrop
= ctl_v1
->headdrop
;
678 q
->limit
= min_t(u32
, ctl
->limit
, q
->maxdepth
* q
->maxflows
);
679 q
->maxflows
= min_t(u32
, q
->maxflows
, q
->limit
);
683 while (sch
->q
.qlen
> q
->limit
) {
684 dropped
+= sfq_drop(sch
, &to_free
);
689 rtnl_kfree_skbs(to_free
, tail
);
690 qdisc_tree_reduce_backlog(sch
, qlen
- sch
->q
.qlen
, dropped
);
692 del_timer(&q
->perturb_timer
);
693 if (q
->perturb_period
) {
694 mod_timer(&q
->perturb_timer
, jiffies
+ q
->perturb_period
);
695 get_random_bytes(&q
->perturbation
, sizeof(q
->perturbation
));
697 sch_tree_unlock(sch
);
702 static void *sfq_alloc(size_t sz
)
704 return kvmalloc(sz
, GFP_KERNEL
);
707 static void sfq_free(void *addr
)
712 static void sfq_destroy(struct Qdisc
*sch
)
714 struct sfq_sched_data
*q
= qdisc_priv(sch
);
716 tcf_block_put(q
->block
);
717 q
->perturb_period
= 0;
718 del_timer_sync(&q
->perturb_timer
);
724 static int sfq_init(struct Qdisc
*sch
, struct nlattr
*opt
)
726 struct sfq_sched_data
*q
= qdisc_priv(sch
);
730 setup_deferrable_timer(&q
->perturb_timer
, sfq_perturbation
,
733 err
= tcf_block_get(&q
->block
, &q
->filter_list
);
737 for (i
= 0; i
< SFQ_MAX_DEPTH
+ 1; i
++) {
738 q
->dep
[i
].next
= i
+ SFQ_MAX_FLOWS
;
739 q
->dep
[i
].prev
= i
+ SFQ_MAX_FLOWS
;
742 q
->limit
= SFQ_MAX_DEPTH
;
743 q
->maxdepth
= SFQ_MAX_DEPTH
;
746 q
->divisor
= SFQ_DEFAULT_HASH_DIVISOR
;
747 q
->maxflows
= SFQ_DEFAULT_FLOWS
;
748 q
->quantum
= psched_mtu(qdisc_dev(sch
));
749 q
->scaled_quantum
= SFQ_ALLOT_SIZE(q
->quantum
);
750 q
->perturb_period
= 0;
751 get_random_bytes(&q
->perturbation
, sizeof(q
->perturbation
));
754 int err
= sfq_change(sch
, opt
);
759 q
->ht
= sfq_alloc(sizeof(q
->ht
[0]) * q
->divisor
);
760 q
->slots
= sfq_alloc(sizeof(q
->slots
[0]) * q
->maxflows
);
761 if (!q
->ht
|| !q
->slots
) {
762 /* Note: sfq_destroy() will be called by our caller */
766 for (i
= 0; i
< q
->divisor
; i
++)
767 q
->ht
[i
] = SFQ_EMPTY_SLOT
;
769 for (i
= 0; i
< q
->maxflows
; i
++) {
770 slot_queue_init(&q
->slots
[i
]);
774 sch
->flags
|= TCQ_F_CAN_BYPASS
;
776 sch
->flags
&= ~TCQ_F_CAN_BYPASS
;
780 static int sfq_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
782 struct sfq_sched_data
*q
= qdisc_priv(sch
);
783 unsigned char *b
= skb_tail_pointer(skb
);
784 struct tc_sfq_qopt_v1 opt
;
785 struct red_parms
*p
= q
->red_parms
;
787 memset(&opt
, 0, sizeof(opt
));
788 opt
.v0
.quantum
= q
->quantum
;
789 opt
.v0
.perturb_period
= q
->perturb_period
/ HZ
;
790 opt
.v0
.limit
= q
->limit
;
791 opt
.v0
.divisor
= q
->divisor
;
792 opt
.v0
.flows
= q
->maxflows
;
793 opt
.depth
= q
->maxdepth
;
794 opt
.headdrop
= q
->headdrop
;
797 opt
.qth_min
= p
->qth_min
>> p
->Wlog
;
798 opt
.qth_max
= p
->qth_max
>> p
->Wlog
;
801 opt
.Scell_log
= p
->Scell_log
;
802 opt
.max_P
= p
->max_P
;
804 memcpy(&opt
.stats
, &q
->stats
, sizeof(opt
.stats
));
805 opt
.flags
= q
->flags
;
807 if (nla_put(skb
, TCA_OPTIONS
, sizeof(opt
), &opt
))
808 goto nla_put_failure
;
817 static struct Qdisc
*sfq_leaf(struct Qdisc
*sch
, unsigned long arg
)
822 static unsigned long sfq_find(struct Qdisc
*sch
, u32 classid
)
827 static unsigned long sfq_bind(struct Qdisc
*sch
, unsigned long parent
,
830 /* we cannot bypass queue discipline anymore */
831 sch
->flags
&= ~TCQ_F_CAN_BYPASS
;
835 static void sfq_unbind(struct Qdisc
*q
, unsigned long cl
)
839 static struct tcf_block
*sfq_tcf_block(struct Qdisc
*sch
, unsigned long cl
)
841 struct sfq_sched_data
*q
= qdisc_priv(sch
);
848 static int sfq_dump_class(struct Qdisc
*sch
, unsigned long cl
,
849 struct sk_buff
*skb
, struct tcmsg
*tcm
)
851 tcm
->tcm_handle
|= TC_H_MIN(cl
);
855 static int sfq_dump_class_stats(struct Qdisc
*sch
, unsigned long cl
,
858 struct sfq_sched_data
*q
= qdisc_priv(sch
);
859 sfq_index idx
= q
->ht
[cl
- 1];
860 struct gnet_stats_queue qs
= { 0 };
861 struct tc_sfq_xstats xstats
= { 0 };
863 if (idx
!= SFQ_EMPTY_SLOT
) {
864 const struct sfq_slot
*slot
= &q
->slots
[idx
];
866 xstats
.allot
= slot
->allot
<< SFQ_ALLOT_SHIFT
;
867 qs
.qlen
= slot
->qlen
;
868 qs
.backlog
= slot
->backlog
;
870 if (gnet_stats_copy_queue(d
, NULL
, &qs
, qs
.qlen
) < 0)
872 return gnet_stats_copy_app(d
, &xstats
, sizeof(xstats
));
875 static void sfq_walk(struct Qdisc
*sch
, struct qdisc_walker
*arg
)
877 struct sfq_sched_data
*q
= qdisc_priv(sch
);
883 for (i
= 0; i
< q
->divisor
; i
++) {
884 if (q
->ht
[i
] == SFQ_EMPTY_SLOT
||
885 arg
->count
< arg
->skip
) {
889 if (arg
->fn(sch
, i
+ 1, arg
) < 0) {
897 static const struct Qdisc_class_ops sfq_class_ops
= {
900 .tcf_block
= sfq_tcf_block
,
901 .bind_tcf
= sfq_bind
,
902 .unbind_tcf
= sfq_unbind
,
903 .dump
= sfq_dump_class
,
904 .dump_stats
= sfq_dump_class_stats
,
908 static struct Qdisc_ops sfq_qdisc_ops __read_mostly
= {
909 .cl_ops
= &sfq_class_ops
,
911 .priv_size
= sizeof(struct sfq_sched_data
),
912 .enqueue
= sfq_enqueue
,
913 .dequeue
= sfq_dequeue
,
914 .peek
= qdisc_peek_dequeued
,
917 .destroy
= sfq_destroy
,
920 .owner
= THIS_MODULE
,
923 static int __init
sfq_module_init(void)
925 return register_qdisc(&sfq_qdisc_ops
);
927 static void __exit
sfq_module_exit(void)
929 unregister_qdisc(&sfq_qdisc_ops
);
931 module_init(sfq_module_init
)
932 module_exit(sfq_module_exit
)
933 MODULE_LICENSE("GPL");