blob: 60d47180f0432e0d65c92ef5a6f6cf865b29a3c0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * net/sched/sch_sfq.c Stochastic Fairness Queueing discipline.
3 *
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.
8 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 */
11
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/types.h>
14#include <linux/kernel.h>
15#include <linux/jiffies.h>
16#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/in.h>
18#include <linux/errno.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/init.h>
Patrick McHardy0ba48052007-07-02 22:49:07 -070020#include <linux/skbuff.h>
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -070021#include <linux/jhash.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090022#include <linux/slab.h>
Eric Dumazet817fb152011-01-20 00:14:58 +000023#include <linux/vmalloc.h>
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -070024#include <net/netlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <net/pkt_sched.h>
Eric Dumazet11fca932011-11-29 03:40:45 +000026#include <net/flow_keys.h>
Eric Dumazetddecf0f2012-01-06 06:31:44 +000027#include <net/red.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29
30/* Stochastic Fairness Queuing algorithm.
31 =======================================
32
33 Source:
34 Paul E. McKenney "Stochastic Fairness Queuing",
35 IEEE INFOCOMM'90 Proceedings, San Francisco, 1990.
36
37 Paul E. McKenney "Stochastic Fairness Queuing",
38 "Interworking: Research and Experience", v.2, 1991, p.113-131.
39
40
41 See also:
42 M. Shreedhar and George Varghese "Efficient Fair
43 Queuing using Deficit Round Robin", Proc. SIGCOMM 95.
44
45
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090046 This is not the thing that is usually called (W)FQ nowadays.
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 It does not use any timestamp mechanism, but instead
48 processes queues in round-robin order.
49
50 ADVANTAGE:
51
52 - It is very cheap. Both CPU and memory requirements are minimal.
53
54 DRAWBACKS:
55
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090056 - "Stochastic" -> It is not 100% fair.
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 When hash collisions occur, several flows are considered as one.
58
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.
64
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.
68
69 IMPLEMENTATION:
Eric Dumazet18cb8092012-01-04 14:18:38 +000070 This implementation limits :
71 - maximal queue length per flow to 127 packets.
72 - max mtu to 2^18-1;
73 - max 65408 flows,
74 - number of hash buckets to 65536.
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76 It is easy to increase these values, but not in flight. */
77
Eric Dumazet18cb8092012-01-04 14:18:38 +000078#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
Eric Dumazet817fb152011-01-20 00:14:58 +000082#define SFQ_DEFAULT_HASH_DIVISOR 1024
83
Eric Dumazeteeaeb062010-12-28 21:53:33 +000084/* 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.
86 */
87#define SFQ_ALLOT_SHIFT 3
88#define SFQ_ALLOT_SIZE(X) DIV_ROUND_UP(X, 1 << SFQ_ALLOT_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Eric Dumazet18cb8092012-01-04 14:18:38 +000090/* This type should contain at least SFQ_MAX_DEPTH + 1 + SFQ_MAX_FLOWS values */
91typedef u16 sfq_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Eric Dumazeteda83e32010-12-20 12:54:58 +000093/*
94 * We dont use pointers to save space.
Eric Dumazet18cb8092012-01-04 14:18:38 +000095 * 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]
Eric Dumazeteda83e32010-12-20 12:54:58 +000097 * are 'pointers' to dep[] array
98 */
Eric Dumazetcc7ec452011-01-19 19:26:56 +000099struct sfq_head {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 sfq_index next;
101 sfq_index prev;
102};
103
Eric Dumazeteda83e32010-12-20 12:54:58 +0000104struct sfq_slot {
105 struct sk_buff *skblist_next;
106 struct sk_buff *skblist_prev;
107 sfq_index qlen; /* number of skbs in skblist */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000108 sfq_index next; /* next slot in sfq RR chain */
Eric Dumazeteda83e32010-12-20 12:54:58 +0000109 struct sfq_head dep; /* anchor in dep[] chains */
110 unsigned short hash; /* hash value (index in ht[]) */
111 short allot; /* credit for this slot */
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000112
113 unsigned int backlog;
114 struct red_vars vars;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000115};
116
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000117struct sfq_sched_data {
Eric Dumazet18cb8092012-01-04 14:18:38 +0000118/* frequently used fields */
119 int limit; /* limit of total number of packets in this qdisc */
Eric Dumazet817fb152011-01-20 00:14:58 +0000120 unsigned int divisor; /* number of slots in hash table */
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000121 u8 headdrop;
122 u8 maxdepth; /* limit of packets per flow */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000123
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700124 u32 perturbation;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000125 u8 cur_depth; /* depth of longest slot */
126 u8 flags;
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000127 unsigned short scaled_quantum; /* SFQ_ALLOT_SIZE(quantum) */
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000128 struct tcf_proto *filter_list;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000129 sfq_index *ht; /* Hash table ('divisor' slots) */
130 struct sfq_slot *slots; /* Flows table ('maxflows' entries) */
131
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000132 struct red_parms *red_parms;
133 struct tc_sfqred_stats stats;
134 struct sfq_slot *tail; /* current slot in round */
135
Eric Dumazet18cb8092012-01-04 14:18:38 +0000136 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
141 */
142
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000143 unsigned int maxflows; /* number of flows in flows array */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000144 int perturb_period;
145 unsigned int quantum; /* Allotment per round: MUST BE >= MTU */
146 struct timer_list perturb_timer;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147};
148
Eric Dumazeteda83e32010-12-20 12:54:58 +0000149/*
150 * sfq_head are either in a sfq_slot or in dep[] array
151 */
152static inline struct sfq_head *sfq_dep_head(struct sfq_sched_data *q, sfq_index val)
153{
Eric Dumazet18cb8092012-01-04 14:18:38 +0000154 if (val < SFQ_MAX_FLOWS)
Eric Dumazeteda83e32010-12-20 12:54:58 +0000155 return &q->slots[val].dep;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000156 return &q->dep[val - SFQ_MAX_FLOWS];
Eric Dumazeteda83e32010-12-20 12:54:58 +0000157}
158
Eric Dumazet225d9b82011-12-21 03:30:11 +0000159/*
160 * In order to be able to quickly rehash our queue when timer changes
161 * q->perturbation, we store flow_keys in skb->cb[]
162 */
163struct sfq_skb_cb {
164 struct flow_keys keys;
165};
166
167static inline struct sfq_skb_cb *sfq_skb_cb(const struct sk_buff *skb)
168{
David S. Miller16bda132012-02-06 15:14:37 -0500169 qdisc_cb_private_validate(skb, sizeof(struct sfq_skb_cb));
170 return (struct sfq_skb_cb *)qdisc_skb_cb(skb)->data;
Eric Dumazet225d9b82011-12-21 03:30:11 +0000171}
172
Eric Dumazet11fca932011-11-29 03:40:45 +0000173static unsigned int sfq_hash(const struct sfq_sched_data *q,
174 const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175{
Eric Dumazet225d9b82011-12-21 03:30:11 +0000176 const struct flow_keys *keys = &sfq_skb_cb(skb)->keys;
Eric Dumazet11fca932011-11-29 03:40:45 +0000177 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Eric Dumazet225d9b82011-12-21 03:30:11 +0000179 hash = jhash_3words((__force u32)keys->dst,
180 (__force u32)keys->src ^ keys->ip_proto,
181 (__force u32)keys->ports, q->perturbation);
Eric Dumazet11fca932011-11-29 03:40:45 +0000182 return hash & (q->divisor - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800185static unsigned int sfq_classify(struct sk_buff *skb, struct Qdisc *sch,
186 int *qerr)
187{
188 struct sfq_sched_data *q = qdisc_priv(sch);
189 struct tcf_result res;
190 int result;
191
192 if (TC_H_MAJ(skb->priority) == sch->handle &&
193 TC_H_MIN(skb->priority) > 0 &&
Eric Dumazet817fb152011-01-20 00:14:58 +0000194 TC_H_MIN(skb->priority) <= q->divisor)
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800195 return TC_H_MIN(skb->priority);
196
Eric Dumazet225d9b82011-12-21 03:30:11 +0000197 if (!q->filter_list) {
198 skb_flow_dissect(skb, &sfq_skb_cb(skb)->keys);
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800199 return sfq_hash(q, skb) + 1;
Eric Dumazet225d9b82011-12-21 03:30:11 +0000200 }
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800201
Jarek Poplawskic27f3392008-08-04 22:39:11 -0700202 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800203 result = tc_classify(skb, q->filter_list, &res);
204 if (result >= 0) {
205#ifdef CONFIG_NET_CLS_ACT
206 switch (result) {
207 case TC_ACT_STOLEN:
208 case TC_ACT_QUEUED:
Jarek Poplawski378a2f02008-08-04 22:31:03 -0700209 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800210 case TC_ACT_SHOT:
211 return 0;
212 }
213#endif
Eric Dumazet817fb152011-01-20 00:14:58 +0000214 if (TC_H_MIN(res.classid) <= q->divisor)
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800215 return TC_H_MIN(res.classid);
216 }
217 return 0;
218}
219
Eric Dumazeteda83e32010-12-20 12:54:58 +0000220/*
Eric Dumazet18cb8092012-01-04 14:18:38 +0000221 * x : slot number [0 .. SFQ_MAX_FLOWS - 1]
Eric Dumazeteda83e32010-12-20 12:54:58 +0000222 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223static inline void sfq_link(struct sfq_sched_data *q, sfq_index x)
224{
225 sfq_index p, n;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000226 struct sfq_slot *slot = &q->slots[x];
227 int qlen = slot->qlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
Eric Dumazet18cb8092012-01-04 14:18:38 +0000229 p = qlen + SFQ_MAX_FLOWS;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000230 n = q->dep[qlen].next;
231
Eric Dumazet18cb8092012-01-04 14:18:38 +0000232 slot->dep.next = n;
233 slot->dep.prev = p;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000234
235 q->dep[qlen].next = x; /* sfq_dep_head(q, p)->next = x */
236 sfq_dep_head(q, n)->prev = x;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237}
238
Eric Dumazeteda83e32010-12-20 12:54:58 +0000239#define sfq_unlink(q, x, n, p) \
240 n = q->slots[x].dep.next; \
241 p = q->slots[x].dep.prev; \
242 sfq_dep_head(q, p)->next = n; \
243 sfq_dep_head(q, n)->prev = p
244
245
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246static inline void sfq_dec(struct sfq_sched_data *q, sfq_index x)
247{
248 sfq_index p, n;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000249 int d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
Eric Dumazeteda83e32010-12-20 12:54:58 +0000251 sfq_unlink(q, x, n, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
Eric Dumazeteda83e32010-12-20 12:54:58 +0000253 d = q->slots[x].qlen--;
254 if (n == p && q->cur_depth == d)
255 q->cur_depth--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 sfq_link(q, x);
257}
258
259static inline void sfq_inc(struct sfq_sched_data *q, sfq_index x)
260{
261 sfq_index p, n;
262 int d;
263
Eric Dumazeteda83e32010-12-20 12:54:58 +0000264 sfq_unlink(q, x, n, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Eric Dumazeteda83e32010-12-20 12:54:58 +0000266 d = ++q->slots[x].qlen;
267 if (q->cur_depth < d)
268 q->cur_depth = d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 sfq_link(q, x);
270}
271
Eric Dumazeteda83e32010-12-20 12:54:58 +0000272/* helper functions : might be changed when/if skb use a standard list_head */
273
274/* remove one skb from tail of slot queue */
275static inline struct sk_buff *slot_dequeue_tail(struct sfq_slot *slot)
276{
277 struct sk_buff *skb = slot->skblist_prev;
278
279 slot->skblist_prev = skb->prev;
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800280 skb->prev->next = (struct sk_buff *)slot;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000281 skb->next = skb->prev = NULL;
282 return skb;
283}
284
285/* remove one skb from head of slot queue */
286static inline struct sk_buff *slot_dequeue_head(struct sfq_slot *slot)
287{
288 struct sk_buff *skb = slot->skblist_next;
289
290 slot->skblist_next = skb->next;
Eric Dumazet18c8d822010-12-31 12:48:55 -0800291 skb->next->prev = (struct sk_buff *)slot;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000292 skb->next = skb->prev = NULL;
293 return skb;
294}
295
296static inline void slot_queue_init(struct sfq_slot *slot)
297{
Eric Dumazet18cb8092012-01-04 14:18:38 +0000298 memset(slot, 0, sizeof(*slot));
Eric Dumazeteda83e32010-12-20 12:54:58 +0000299 slot->skblist_prev = slot->skblist_next = (struct sk_buff *)slot;
300}
301
302/* add skb to slot queue (tail add) */
303static inline void slot_queue_add(struct sfq_slot *slot, struct sk_buff *skb)
304{
305 skb->prev = slot->skblist_prev;
306 skb->next = (struct sk_buff *)slot;
307 slot->skblist_prev->next = skb;
308 slot->skblist_prev = skb;
309}
310
311#define slot_queue_walk(slot, skb) \
312 for (skb = slot->skblist_next; \
313 skb != (struct sk_buff *)slot; \
314 skb = skb->next)
315
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316static unsigned int sfq_drop(struct Qdisc *sch)
317{
318 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazeteda83e32010-12-20 12:54:58 +0000319 sfq_index x, d = q->cur_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 struct sk_buff *skb;
321 unsigned int len;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000322 struct sfq_slot *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Eric Dumazeteda83e32010-12-20 12:54:58 +0000324 /* Queue is full! Find the longest slot and drop tail packet from it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 if (d > 1) {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000326 x = q->dep[d].next;
327 slot = &q->slots[x];
328drop:
Eric Dumazet18cb8092012-01-04 14:18:38 +0000329 skb = q->headdrop ? slot_dequeue_head(slot) : slot_dequeue_tail(slot);
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700330 len = qdisc_pkt_len(skb);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000331 slot->backlog -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 sfq_dec(q, x);
Eric Dumazeteda83e32010-12-20 12:54:58 +0000333 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 sch->q.qlen--;
335 sch->qstats.drops++;
Patrick McHardyf5539eb2006-03-20 19:01:38 -0800336 sch->qstats.backlog -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 return len;
338 }
339
340 if (d == 1) {
341 /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
Eric Dumazeteda83e32010-12-20 12:54:58 +0000342 x = q->tail->next;
343 slot = &q->slots[x];
344 q->tail->next = slot->next;
345 q->ht[slot->hash] = SFQ_EMPTY_SLOT;
346 goto drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 }
348
349 return 0;
350}
351
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000352/* Is ECN parameter configured */
353static int sfq_prob_mark(const struct sfq_sched_data *q)
354{
355 return q->flags & TC_RED_ECN;
356}
357
358/* Should packets over max threshold just be marked */
359static int sfq_hard_mark(const struct sfq_sched_data *q)
360{
361 return (q->flags & (TC_RED_ECN | TC_RED_HARDDROP)) == TC_RED_ECN;
362}
363
364static int sfq_headdrop(const struct sfq_sched_data *q)
365{
366 return q->headdrop;
367}
368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369static int
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800370sfq_enqueue(struct sk_buff *skb, struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
372 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800373 unsigned int hash;
Eric Dumazet8efa8852011-05-23 11:02:42 +0000374 sfq_index x, qlen;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000375 struct sfq_slot *slot;
Jarek Poplawski7f3ff4f2008-12-21 20:14:48 -0800376 int uninitialized_var(ret);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000377 struct sk_buff *head;
378 int delta;
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800379
380 hash = sfq_classify(skb, sch, &ret);
381 if (hash == 0) {
Jarek Poplawskic27f3392008-08-04 22:39:11 -0700382 if (ret & __NET_XMIT_BYPASS)
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800383 sch->qstats.drops++;
384 kfree_skb(skb);
385 return ret;
386 }
387 hash--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389 x = q->ht[hash];
Eric Dumazeteda83e32010-12-20 12:54:58 +0000390 slot = &q->slots[x];
391 if (x == SFQ_EMPTY_SLOT) {
392 x = q->dep[0].next; /* get a free slot */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000393 if (x >= SFQ_MAX_FLOWS)
394 return qdisc_drop(skb, sch);
Eric Dumazeteda83e32010-12-20 12:54:58 +0000395 q->ht[hash] = x;
396 slot = &q->slots[x];
397 slot->hash = hash;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000398 slot->backlog = 0; /* should already be 0 anyway... */
399 red_set_vars(&slot->vars);
400 goto enqueue;
401 }
402 if (q->red_parms) {
403 slot->vars.qavg = red_calc_qavg_no_idle_time(q->red_parms,
404 &slot->vars,
405 slot->backlog);
406 switch (red_action(q->red_parms,
407 &slot->vars,
408 slot->vars.qavg)) {
409 case RED_DONT_MARK:
410 break;
411
412 case RED_PROB_MARK:
413 sch->qstats.overlimits++;
414 if (sfq_prob_mark(q)) {
415 /* We know we have at least one packet in queue */
416 if (sfq_headdrop(q) &&
417 INET_ECN_set_ce(slot->skblist_next)) {
418 q->stats.prob_mark_head++;
419 break;
420 }
421 if (INET_ECN_set_ce(skb)) {
422 q->stats.prob_mark++;
423 break;
424 }
425 }
426 q->stats.prob_drop++;
427 goto congestion_drop;
428
429 case RED_HARD_MARK:
430 sch->qstats.overlimits++;
431 if (sfq_hard_mark(q)) {
432 /* We know we have at least one packet in queue */
433 if (sfq_headdrop(q) &&
434 INET_ECN_set_ce(slot->skblist_next)) {
435 q->stats.forced_mark_head++;
436 break;
437 }
438 if (INET_ECN_set_ce(skb)) {
439 q->stats.forced_mark++;
440 break;
441 }
442 }
443 q->stats.forced_drop++;
444 goto congestion_drop;
445 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800447
Eric Dumazet18cb8092012-01-04 14:18:38 +0000448 if (slot->qlen >= q->maxdepth) {
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000449congestion_drop:
450 if (!sfq_headdrop(q))
Eric Dumazet18cb8092012-01-04 14:18:38 +0000451 return qdisc_drop(skb, sch);
452
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000453 /* We know we have at least one packet in queue */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000454 head = slot_dequeue_head(slot);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000455 delta = qdisc_pkt_len(head) - qdisc_pkt_len(skb);
456 sch->qstats.backlog -= delta;
457 slot->backlog -= delta;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000458 qdisc_drop(head, sch);
459
Eric Dumazet18cb8092012-01-04 14:18:38 +0000460 slot_queue_add(slot, skb);
461 return NET_XMIT_CN;
462 }
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700463
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000464enqueue:
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700465 sch->qstats.backlog += qdisc_pkt_len(skb);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000466 slot->backlog += qdisc_pkt_len(skb);
Eric Dumazeteda83e32010-12-20 12:54:58 +0000467 slot_queue_add(slot, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 sfq_inc(q, x);
Eric Dumazeteda83e32010-12-20 12:54:58 +0000469 if (slot->qlen == 1) { /* The flow is new */
470 if (q->tail == NULL) { /* It is the first flow */
471 slot->next = x;
Eric Dumazetd47a0ac2012-01-01 18:33:31 +0000472 q->tail = slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 } else {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000474 slot->next = q->tail->next;
475 q->tail->next = x;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 }
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000477 /* We could use a bigger initial quantum for new flows */
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000478 slot->allot = q->scaled_quantum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 }
Eric Dumazet9190b3b2011-01-20 23:31:33 -0800480 if (++sch->q.qlen <= q->limit)
Ben Greear9871e502010-08-10 01:45:40 -0700481 return NET_XMIT_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
Eric Dumazet8efa8852011-05-23 11:02:42 +0000483 qlen = slot->qlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 sfq_drop(sch);
Eric Dumazet8efa8852011-05-23 11:02:42 +0000485 /* Return Congestion Notification only if we dropped a packet
486 * from this flow.
487 */
Eric Dumazete1738bd2011-07-29 19:22:42 +0000488 if (qlen != slot->qlen)
489 return NET_XMIT_CN;
490
491 /* As we dropped a packet, better let upper stack know this */
492 qdisc_tree_decrease_qlen(sch, 1);
493 return NET_XMIT_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494}
495
Patrick McHardy48a8f512008-10-31 00:44:18 -0700496static struct sk_buff *
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800497sfq_dequeue(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498{
499 struct sfq_sched_data *q = qdisc_priv(sch);
500 struct sk_buff *skb;
Eric Dumazetaa3e2192010-12-20 13:18:16 -0800501 sfq_index a, next_a;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000502 struct sfq_slot *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504 /* No active slots */
Eric Dumazeteda83e32010-12-20 12:54:58 +0000505 if (q->tail == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 return NULL;
507
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000508next_slot:
Eric Dumazeteda83e32010-12-20 12:54:58 +0000509 a = q->tail->next;
510 slot = &q->slots[a];
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000511 if (slot->allot <= 0) {
512 q->tail = slot;
513 slot->allot += q->scaled_quantum;
514 goto next_slot;
515 }
Eric Dumazeteda83e32010-12-20 12:54:58 +0000516 skb = slot_dequeue_head(slot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 sfq_dec(q, a);
Eric Dumazet9190b3b2011-01-20 23:31:33 -0800518 qdisc_bstats_update(sch, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 sch->q.qlen--;
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700520 sch->qstats.backlog -= qdisc_pkt_len(skb);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000521 slot->backlog -= qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 /* Is the slot empty? */
Eric Dumazeteda83e32010-12-20 12:54:58 +0000523 if (slot->qlen == 0) {
524 q->ht[slot->hash] = SFQ_EMPTY_SLOT;
525 next_a = slot->next;
Eric Dumazetaa3e2192010-12-20 13:18:16 -0800526 if (a == next_a) {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000527 q->tail = NULL; /* no more active slots */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 return skb;
529 }
Eric Dumazeteda83e32010-12-20 12:54:58 +0000530 q->tail->next = next_a;
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000531 } else {
532 slot->allot -= SFQ_ALLOT_SIZE(qdisc_pkt_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
534 return skb;
535}
536
537static void
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800538sfq_reset(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
540 struct sk_buff *skb;
541
542 while ((skb = sfq_dequeue(sch)) != NULL)
543 kfree_skb(skb);
544}
545
Eric Dumazet225d9b82011-12-21 03:30:11 +0000546/*
547 * When q->perturbation is changed, we rehash all queued skbs
548 * to avoid OOO (Out Of Order) effects.
549 * We dont use sfq_dequeue()/sfq_enqueue() because we dont want to change
550 * counters.
551 */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000552static void sfq_rehash(struct Qdisc *sch)
Eric Dumazet225d9b82011-12-21 03:30:11 +0000553{
Eric Dumazet18cb8092012-01-04 14:18:38 +0000554 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000555 struct sk_buff *skb;
556 int i;
557 struct sfq_slot *slot;
558 struct sk_buff_head list;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000559 int dropped = 0;
Eric Dumazet225d9b82011-12-21 03:30:11 +0000560
561 __skb_queue_head_init(&list);
562
Eric Dumazet18cb8092012-01-04 14:18:38 +0000563 for (i = 0; i < q->maxflows; i++) {
Eric Dumazet225d9b82011-12-21 03:30:11 +0000564 slot = &q->slots[i];
565 if (!slot->qlen)
566 continue;
567 while (slot->qlen) {
568 skb = slot_dequeue_head(slot);
569 sfq_dec(q, i);
570 __skb_queue_tail(&list, skb);
571 }
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000572 slot->backlog = 0;
573 red_set_vars(&slot->vars);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000574 q->ht[slot->hash] = SFQ_EMPTY_SLOT;
575 }
576 q->tail = NULL;
577
578 while ((skb = __skb_dequeue(&list)) != NULL) {
579 unsigned int hash = sfq_hash(q, skb);
580 sfq_index x = q->ht[hash];
581
582 slot = &q->slots[x];
583 if (x == SFQ_EMPTY_SLOT) {
584 x = q->dep[0].next; /* get a free slot */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000585 if (x >= SFQ_MAX_FLOWS) {
586drop: sch->qstats.backlog -= qdisc_pkt_len(skb);
587 kfree_skb(skb);
588 dropped++;
589 continue;
590 }
Eric Dumazet225d9b82011-12-21 03:30:11 +0000591 q->ht[hash] = x;
592 slot = &q->slots[x];
593 slot->hash = hash;
594 }
Eric Dumazet18cb8092012-01-04 14:18:38 +0000595 if (slot->qlen >= q->maxdepth)
596 goto drop;
Eric Dumazet225d9b82011-12-21 03:30:11 +0000597 slot_queue_add(slot, skb);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000598 if (q->red_parms)
599 slot->vars.qavg = red_calc_qavg(q->red_parms,
600 &slot->vars,
601 slot->backlog);
602 slot->backlog += qdisc_pkt_len(skb);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000603 sfq_inc(q, x);
604 if (slot->qlen == 1) { /* The flow is new */
605 if (q->tail == NULL) { /* It is the first flow */
606 slot->next = x;
607 } else {
608 slot->next = q->tail->next;
609 q->tail->next = x;
610 }
611 q->tail = slot;
612 slot->allot = q->scaled_quantum;
613 }
614 }
Eric Dumazet18cb8092012-01-04 14:18:38 +0000615 sch->q.qlen -= dropped;
616 qdisc_tree_decrease_qlen(sch, dropped);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000617}
618
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619static void sfq_perturbation(unsigned long arg)
620{
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800621 struct Qdisc *sch = (struct Qdisc *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000623 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Eric Dumazet225d9b82011-12-21 03:30:11 +0000625 spin_lock(root_lock);
Stephen Hemmingerd46f8dd2008-01-20 17:19:43 -0800626 q->perturbation = net_random();
Eric Dumazet225d9b82011-12-21 03:30:11 +0000627 if (!q->filter_list && q->tail)
Eric Dumazet18cb8092012-01-04 14:18:38 +0000628 sfq_rehash(sch);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000629 spin_unlock(root_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700631 if (q->perturb_period)
632 mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633}
634
Patrick McHardy1e904742008-01-22 22:11:17 -0800635static int sfq_change(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
637 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800638 struct tc_sfq_qopt *ctl = nla_data(opt);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000639 struct tc_sfq_qopt_v1 *ctl_v1 = NULL;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800640 unsigned int qlen;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000641 struct red_parms *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Patrick McHardy1e904742008-01-22 22:11:17 -0800643 if (opt->nla_len < nla_attr_size(sizeof(*ctl)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 return -EINVAL;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000645 if (opt->nla_len >= nla_attr_size(sizeof(*ctl_v1)))
646 ctl_v1 = nla_data(opt);
stephen hemminger119b3d32011-02-02 15:19:51 +0000647 if (ctl->divisor &&
648 (!is_power_of_2(ctl->divisor) || ctl->divisor > 65536))
649 return -EINVAL;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000650 if (ctl_v1 && ctl_v1->qth_min) {
651 p = kmalloc(sizeof(*p), GFP_KERNEL);
652 if (!p)
653 return -ENOMEM;
654 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 sch_tree_lock(sch);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000656 if (ctl->quantum) {
657 q->quantum = ctl->quantum;
658 q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum);
659 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800660 q->perturb_period = ctl->perturb_period * HZ;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000661 if (ctl->flows)
662 q->maxflows = min_t(u32, ctl->flows, SFQ_MAX_FLOWS);
663 if (ctl->divisor) {
Eric Dumazet817fb152011-01-20 00:14:58 +0000664 q->divisor = ctl->divisor;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000665 q->maxflows = min_t(u32, q->maxflows, q->divisor);
666 }
667 if (ctl_v1) {
668 if (ctl_v1->depth)
669 q->maxdepth = min_t(u32, ctl_v1->depth, SFQ_MAX_DEPTH);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000670 if (p) {
671 swap(q->red_parms, p);
672 red_set_parms(q->red_parms,
673 ctl_v1->qth_min, ctl_v1->qth_max,
674 ctl_v1->Wlog,
675 ctl_v1->Plog, ctl_v1->Scell_log,
676 NULL,
677 ctl_v1->max_P);
678 }
679 q->flags = ctl_v1->flags;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000680 q->headdrop = ctl_v1->headdrop;
681 }
682 if (ctl->limit) {
683 q->limit = min_t(u32, ctl->limit, q->maxdepth * q->maxflows);
684 q->maxflows = min_t(u32, q->maxflows, q->limit);
685 }
686
Patrick McHardy5e50da02006-11-29 17:36:20 -0800687 qlen = sch->q.qlen;
Alexey Kuznetsov5588b402007-09-19 10:42:03 -0700688 while (sch->q.qlen > q->limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 sfq_drop(sch);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800690 qdisc_tree_decrease_qlen(sch, qlen - sch->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
692 del_timer(&q->perturb_timer);
693 if (q->perturb_period) {
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700694 mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
Stephen Hemmingerd46f8dd2008-01-20 17:19:43 -0800695 q->perturbation = net_random();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 }
697 sch_tree_unlock(sch);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000698 kfree(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 return 0;
700}
701
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000702static void *sfq_alloc(size_t sz)
703{
704 void *ptr = kmalloc(sz, GFP_KERNEL | __GFP_NOWARN);
705
706 if (!ptr)
707 ptr = vmalloc(sz);
708 return ptr;
709}
710
711static void sfq_free(void *addr)
712{
713 if (addr) {
714 if (is_vmalloc_addr(addr))
715 vfree(addr);
716 else
717 kfree(addr);
718 }
719}
720
721static void sfq_destroy(struct Qdisc *sch)
722{
723 struct sfq_sched_data *q = qdisc_priv(sch);
724
725 tcf_destroy_chain(&q->filter_list);
726 q->perturb_period = 0;
727 del_timer_sync(&q->perturb_timer);
728 sfq_free(q->ht);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000729 sfq_free(q->slots);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000730 kfree(q->red_parms);
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000731}
732
Patrick McHardy1e904742008-01-22 22:11:17 -0800733static int sfq_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734{
735 struct sfq_sched_data *q = qdisc_priv(sch);
736 int i;
737
Stephen Hemmingerd3e99482008-01-20 17:18:45 -0800738 q->perturb_timer.function = sfq_perturbation;
Fernando Carrijoc19a28e2009-01-07 18:09:08 -0800739 q->perturb_timer.data = (unsigned long)sch;
Stephen Hemmingerd3e99482008-01-20 17:18:45 -0800740 init_timer_deferrable(&q->perturb_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Eric Dumazet18cb8092012-01-04 14:18:38 +0000742 for (i = 0; i < SFQ_MAX_DEPTH + 1; i++) {
743 q->dep[i].next = i + SFQ_MAX_FLOWS;
744 q->dep[i].prev = i + SFQ_MAX_FLOWS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800746
Eric Dumazet18cb8092012-01-04 14:18:38 +0000747 q->limit = SFQ_MAX_DEPTH;
748 q->maxdepth = SFQ_MAX_DEPTH;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000749 q->cur_depth = 0;
750 q->tail = NULL;
Eric Dumazet817fb152011-01-20 00:14:58 +0000751 q->divisor = SFQ_DEFAULT_HASH_DIVISOR;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000752 q->maxflows = SFQ_DEFAULT_FLOWS;
Eric Dumazet02a90982012-01-04 06:23:01 +0000753 q->quantum = psched_mtu(qdisc_dev(sch));
754 q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum);
755 q->perturb_period = 0;
756 q->perturbation = net_random();
757
758 if (opt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 int err = sfq_change(sch, opt);
760 if (err)
761 return err;
762 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800763
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000764 q->ht = sfq_alloc(sizeof(q->ht[0]) * q->divisor);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000765 q->slots = sfq_alloc(sizeof(q->slots[0]) * q->maxflows);
766 if (!q->ht || !q->slots) {
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000767 sfq_destroy(sch);
Eric Dumazet817fb152011-01-20 00:14:58 +0000768 return -ENOMEM;
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000769 }
Eric Dumazet817fb152011-01-20 00:14:58 +0000770 for (i = 0; i < q->divisor; i++)
771 q->ht[i] = SFQ_EMPTY_SLOT;
772
Eric Dumazet18cb8092012-01-04 14:18:38 +0000773 for (i = 0; i < q->maxflows; i++) {
Eric Dumazet18c8d822010-12-31 12:48:55 -0800774 slot_queue_init(&q->slots[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 sfq_link(q, i);
Eric Dumazet18c8d822010-12-31 12:48:55 -0800776 }
Eric Dumazet23624932011-01-21 16:26:09 -0800777 if (q->limit >= 1)
778 sch->flags |= TCQ_F_CAN_BYPASS;
779 else
780 sch->flags &= ~TCQ_F_CAN_BYPASS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 return 0;
782}
783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784static int sfq_dump(struct Qdisc *sch, struct sk_buff *skb)
785{
786 struct sfq_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700787 unsigned char *b = skb_tail_pointer(skb);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000788 struct tc_sfq_qopt_v1 opt;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000789 struct red_parms *p = q->red_parms;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Eric Dumazet18cb8092012-01-04 14:18:38 +0000791 memset(&opt, 0, sizeof(opt));
792 opt.v0.quantum = q->quantum;
793 opt.v0.perturb_period = q->perturb_period / HZ;
794 opt.v0.limit = q->limit;
795 opt.v0.divisor = q->divisor;
796 opt.v0.flows = q->maxflows;
797 opt.depth = q->maxdepth;
798 opt.headdrop = q->headdrop;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000800 if (p) {
801 opt.qth_min = p->qth_min >> p->Wlog;
802 opt.qth_max = p->qth_max >> p->Wlog;
803 opt.Wlog = p->Wlog;
804 opt.Plog = p->Plog;
805 opt.Scell_log = p->Scell_log;
806 opt.max_P = p->max_P;
807 }
808 memcpy(&opt.stats, &q->stats, sizeof(opt.stats));
809 opt.flags = q->flags;
810
Patrick McHardy1e904742008-01-22 22:11:17 -0800811 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
813 return skb->len;
814
Patrick McHardy1e904742008-01-22 22:11:17 -0800815nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700816 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 return -1;
818}
819
Jarek Poplawski41065fb2010-08-10 22:31:02 +0000820static struct Qdisc *sfq_leaf(struct Qdisc *sch, unsigned long arg)
821{
822 return NULL;
823}
824
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800825static unsigned long sfq_get(struct Qdisc *sch, u32 classid)
826{
827 return 0;
828}
829
Jarek Poplawskieb4a5522010-08-06 00:22:35 +0000830static unsigned long sfq_bind(struct Qdisc *sch, unsigned long parent,
831 u32 classid)
832{
Eric Dumazet23624932011-01-21 16:26:09 -0800833 /* we cannot bypass queue discipline anymore */
834 sch->flags &= ~TCQ_F_CAN_BYPASS;
Jarek Poplawskieb4a5522010-08-06 00:22:35 +0000835 return 0;
836}
837
Jarek Poplawskida7115d2010-08-09 12:18:17 +0000838static void sfq_put(struct Qdisc *q, unsigned long cl)
839{
840}
841
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800842static struct tcf_proto **sfq_find_tcf(struct Qdisc *sch, unsigned long cl)
843{
844 struct sfq_sched_data *q = qdisc_priv(sch);
845
846 if (cl)
847 return NULL;
848 return &q->filter_list;
849}
850
Patrick McHardy94de78d2008-01-31 18:37:16 -0800851static int sfq_dump_class(struct Qdisc *sch, unsigned long cl,
852 struct sk_buff *skb, struct tcmsg *tcm)
853{
854 tcm->tcm_handle |= TC_H_MIN(cl);
855 return 0;
856}
857
858static int sfq_dump_class_stats(struct Qdisc *sch, unsigned long cl,
859 struct gnet_dump *d)
860{
861 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800862 sfq_index idx = q->ht[cl - 1];
863 struct gnet_stats_queue qs = { 0 };
864 struct tc_sfq_xstats xstats = { 0 };
Eric Dumazetc4266262010-12-15 08:18:36 +0000865
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800866 if (idx != SFQ_EMPTY_SLOT) {
867 const struct sfq_slot *slot = &q->slots[idx];
Patrick McHardy94de78d2008-01-31 18:37:16 -0800868
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000869 xstats.allot = slot->allot << SFQ_ALLOT_SHIFT;
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800870 qs.qlen = slot->qlen;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000871 qs.backlog = slot->backlog;
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800872 }
Patrick McHardy94de78d2008-01-31 18:37:16 -0800873 if (gnet_stats_copy_queue(d, &qs) < 0)
874 return -1;
875 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
876}
877
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800878static void sfq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
879{
Patrick McHardy94de78d2008-01-31 18:37:16 -0800880 struct sfq_sched_data *q = qdisc_priv(sch);
881 unsigned int i;
882
883 if (arg->stop)
884 return;
885
Eric Dumazet817fb152011-01-20 00:14:58 +0000886 for (i = 0; i < q->divisor; i++) {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000887 if (q->ht[i] == SFQ_EMPTY_SLOT ||
Patrick McHardy94de78d2008-01-31 18:37:16 -0800888 arg->count < arg->skip) {
889 arg->count++;
890 continue;
891 }
892 if (arg->fn(sch, i + 1, arg) < 0) {
893 arg->stop = 1;
894 break;
895 }
896 arg->count++;
897 }
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800898}
899
900static const struct Qdisc_class_ops sfq_class_ops = {
Jarek Poplawski41065fb2010-08-10 22:31:02 +0000901 .leaf = sfq_leaf,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800902 .get = sfq_get,
Jarek Poplawskida7115d2010-08-09 12:18:17 +0000903 .put = sfq_put,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800904 .tcf_chain = sfq_find_tcf,
Jarek Poplawskieb4a5522010-08-06 00:22:35 +0000905 .bind_tcf = sfq_bind,
Jarek Poplawskida7115d2010-08-09 12:18:17 +0000906 .unbind_tcf = sfq_put,
Patrick McHardy94de78d2008-01-31 18:37:16 -0800907 .dump = sfq_dump_class,
908 .dump_stats = sfq_dump_class_stats,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800909 .walk = sfq_walk,
910};
911
Eric Dumazet20fea082007-11-14 01:44:41 -0800912static struct Qdisc_ops sfq_qdisc_ops __read_mostly = {
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800913 .cl_ops = &sfq_class_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 .id = "sfq",
915 .priv_size = sizeof(struct sfq_sched_data),
916 .enqueue = sfq_enqueue,
917 .dequeue = sfq_dequeue,
Eric Dumazet07bd8df2011-05-25 04:40:11 +0000918 .peek = qdisc_peek_dequeued,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 .drop = sfq_drop,
920 .init = sfq_init,
921 .reset = sfq_reset,
922 .destroy = sfq_destroy,
923 .change = NULL,
924 .dump = sfq_dump,
925 .owner = THIS_MODULE,
926};
927
928static int __init sfq_module_init(void)
929{
930 return register_qdisc(&sfq_qdisc_ops);
931}
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900932static void __exit sfq_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933{
934 unregister_qdisc(&sfq_qdisc_ops);
935}
936module_init(sfq_module_init)
937module_exit(sfq_module_exit)
938MODULE_LICENSE("GPL");