blob: 02a21abea65e20c207fe77d81f0240957c259970 [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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 } else {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000473 slot->next = q->tail->next;
474 q->tail->next = x;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 }
Eric Dumazetcc34eb62012-03-13 18:04:25 +0000476 /* We put this flow at the end of our flow list.
477 * This might sound unfair for a new flow to wait after old ones,
478 * but we could endup servicing new flows only, and freeze old ones.
479 */
480 q->tail = slot;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000481 /* We could use a bigger initial quantum for new flows */
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000482 slot->allot = q->scaled_quantum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 }
Eric Dumazet9190b3b2011-01-20 23:31:33 -0800484 if (++sch->q.qlen <= q->limit)
Ben Greear9871e502010-08-10 01:45:40 -0700485 return NET_XMIT_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Eric Dumazet8efa8852011-05-23 11:02:42 +0000487 qlen = slot->qlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 sfq_drop(sch);
Eric Dumazet8efa8852011-05-23 11:02:42 +0000489 /* Return Congestion Notification only if we dropped a packet
490 * from this flow.
491 */
Eric Dumazete1738bd2011-07-29 19:22:42 +0000492 if (qlen != slot->qlen)
493 return NET_XMIT_CN;
494
495 /* As we dropped a packet, better let upper stack know this */
496 qdisc_tree_decrease_qlen(sch, 1);
497 return NET_XMIT_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498}
499
Patrick McHardy48a8f512008-10-31 00:44:18 -0700500static struct sk_buff *
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800501sfq_dequeue(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
503 struct sfq_sched_data *q = qdisc_priv(sch);
504 struct sk_buff *skb;
Eric Dumazetaa3e2192010-12-20 13:18:16 -0800505 sfq_index a, next_a;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000506 struct sfq_slot *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
508 /* No active slots */
Eric Dumazeteda83e32010-12-20 12:54:58 +0000509 if (q->tail == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 return NULL;
511
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000512next_slot:
Eric Dumazeteda83e32010-12-20 12:54:58 +0000513 a = q->tail->next;
514 slot = &q->slots[a];
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000515 if (slot->allot <= 0) {
516 q->tail = slot;
517 slot->allot += q->scaled_quantum;
518 goto next_slot;
519 }
Eric Dumazeteda83e32010-12-20 12:54:58 +0000520 skb = slot_dequeue_head(slot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 sfq_dec(q, a);
Eric Dumazet9190b3b2011-01-20 23:31:33 -0800522 qdisc_bstats_update(sch, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 sch->q.qlen--;
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700524 sch->qstats.backlog -= qdisc_pkt_len(skb);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000525 slot->backlog -= qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 /* Is the slot empty? */
Eric Dumazeteda83e32010-12-20 12:54:58 +0000527 if (slot->qlen == 0) {
528 q->ht[slot->hash] = SFQ_EMPTY_SLOT;
529 next_a = slot->next;
Eric Dumazetaa3e2192010-12-20 13:18:16 -0800530 if (a == next_a) {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000531 q->tail = NULL; /* no more active slots */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 return skb;
533 }
Eric Dumazeteda83e32010-12-20 12:54:58 +0000534 q->tail->next = next_a;
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000535 } else {
536 slot->allot -= SFQ_ALLOT_SIZE(qdisc_pkt_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 }
538 return skb;
539}
540
541static void
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800542sfq_reset(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543{
544 struct sk_buff *skb;
545
546 while ((skb = sfq_dequeue(sch)) != NULL)
547 kfree_skb(skb);
548}
549
Eric Dumazet225d9b82011-12-21 03:30:11 +0000550/*
551 * When q->perturbation is changed, we rehash all queued skbs
552 * to avoid OOO (Out Of Order) effects.
553 * We dont use sfq_dequeue()/sfq_enqueue() because we dont want to change
554 * counters.
555 */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000556static void sfq_rehash(struct Qdisc *sch)
Eric Dumazet225d9b82011-12-21 03:30:11 +0000557{
Eric Dumazet18cb8092012-01-04 14:18:38 +0000558 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000559 struct sk_buff *skb;
560 int i;
561 struct sfq_slot *slot;
562 struct sk_buff_head list;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000563 int dropped = 0;
Eric Dumazet225d9b82011-12-21 03:30:11 +0000564
565 __skb_queue_head_init(&list);
566
Eric Dumazet18cb8092012-01-04 14:18:38 +0000567 for (i = 0; i < q->maxflows; i++) {
Eric Dumazet225d9b82011-12-21 03:30:11 +0000568 slot = &q->slots[i];
569 if (!slot->qlen)
570 continue;
571 while (slot->qlen) {
572 skb = slot_dequeue_head(slot);
573 sfq_dec(q, i);
574 __skb_queue_tail(&list, skb);
575 }
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000576 slot->backlog = 0;
577 red_set_vars(&slot->vars);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000578 q->ht[slot->hash] = SFQ_EMPTY_SLOT;
579 }
580 q->tail = NULL;
581
582 while ((skb = __skb_dequeue(&list)) != NULL) {
583 unsigned int hash = sfq_hash(q, skb);
584 sfq_index x = q->ht[hash];
585
586 slot = &q->slots[x];
587 if (x == SFQ_EMPTY_SLOT) {
588 x = q->dep[0].next; /* get a free slot */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000589 if (x >= SFQ_MAX_FLOWS) {
590drop: sch->qstats.backlog -= qdisc_pkt_len(skb);
591 kfree_skb(skb);
592 dropped++;
593 continue;
594 }
Eric Dumazet225d9b82011-12-21 03:30:11 +0000595 q->ht[hash] = x;
596 slot = &q->slots[x];
597 slot->hash = hash;
598 }
Eric Dumazet18cb8092012-01-04 14:18:38 +0000599 if (slot->qlen >= q->maxdepth)
600 goto drop;
Eric Dumazet225d9b82011-12-21 03:30:11 +0000601 slot_queue_add(slot, skb);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000602 if (q->red_parms)
603 slot->vars.qavg = red_calc_qavg(q->red_parms,
604 &slot->vars,
605 slot->backlog);
606 slot->backlog += qdisc_pkt_len(skb);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000607 sfq_inc(q, x);
608 if (slot->qlen == 1) { /* The flow is new */
609 if (q->tail == NULL) { /* It is the first flow */
610 slot->next = x;
611 } else {
612 slot->next = q->tail->next;
613 q->tail->next = x;
614 }
615 q->tail = slot;
616 slot->allot = q->scaled_quantum;
617 }
618 }
Eric Dumazet18cb8092012-01-04 14:18:38 +0000619 sch->q.qlen -= dropped;
620 qdisc_tree_decrease_qlen(sch, dropped);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000621}
622
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623static void sfq_perturbation(unsigned long arg)
624{
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800625 struct Qdisc *sch = (struct Qdisc *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000627 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Eric Dumazet225d9b82011-12-21 03:30:11 +0000629 spin_lock(root_lock);
Stephen Hemmingerd46f8dd2008-01-20 17:19:43 -0800630 q->perturbation = net_random();
Eric Dumazet225d9b82011-12-21 03:30:11 +0000631 if (!q->filter_list && q->tail)
Eric Dumazet18cb8092012-01-04 14:18:38 +0000632 sfq_rehash(sch);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000633 spin_unlock(root_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700635 if (q->perturb_period)
636 mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
638
Patrick McHardy1e904742008-01-22 22:11:17 -0800639static int sfq_change(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
641 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800642 struct tc_sfq_qopt *ctl = nla_data(opt);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000643 struct tc_sfq_qopt_v1 *ctl_v1 = NULL;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800644 unsigned int qlen;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000645 struct red_parms *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Patrick McHardy1e904742008-01-22 22:11:17 -0800647 if (opt->nla_len < nla_attr_size(sizeof(*ctl)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 return -EINVAL;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000649 if (opt->nla_len >= nla_attr_size(sizeof(*ctl_v1)))
650 ctl_v1 = nla_data(opt);
stephen hemminger119b3d32011-02-02 15:19:51 +0000651 if (ctl->divisor &&
652 (!is_power_of_2(ctl->divisor) || ctl->divisor > 65536))
653 return -EINVAL;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000654 if (ctl_v1 && ctl_v1->qth_min) {
655 p = kmalloc(sizeof(*p), GFP_KERNEL);
656 if (!p)
657 return -ENOMEM;
658 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 sch_tree_lock(sch);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000660 if (ctl->quantum) {
661 q->quantum = ctl->quantum;
662 q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum);
663 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800664 q->perturb_period = ctl->perturb_period * HZ;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000665 if (ctl->flows)
666 q->maxflows = min_t(u32, ctl->flows, SFQ_MAX_FLOWS);
667 if (ctl->divisor) {
Eric Dumazet817fb152011-01-20 00:14:58 +0000668 q->divisor = ctl->divisor;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000669 q->maxflows = min_t(u32, q->maxflows, q->divisor);
670 }
671 if (ctl_v1) {
672 if (ctl_v1->depth)
673 q->maxdepth = min_t(u32, ctl_v1->depth, SFQ_MAX_DEPTH);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000674 if (p) {
675 swap(q->red_parms, p);
676 red_set_parms(q->red_parms,
677 ctl_v1->qth_min, ctl_v1->qth_max,
678 ctl_v1->Wlog,
679 ctl_v1->Plog, ctl_v1->Scell_log,
680 NULL,
681 ctl_v1->max_P);
682 }
683 q->flags = ctl_v1->flags;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000684 q->headdrop = ctl_v1->headdrop;
685 }
686 if (ctl->limit) {
687 q->limit = min_t(u32, ctl->limit, q->maxdepth * q->maxflows);
688 q->maxflows = min_t(u32, q->maxflows, q->limit);
689 }
690
Patrick McHardy5e50da02006-11-29 17:36:20 -0800691 qlen = sch->q.qlen;
Alexey Kuznetsov5588b402007-09-19 10:42:03 -0700692 while (sch->q.qlen > q->limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 sfq_drop(sch);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800694 qdisc_tree_decrease_qlen(sch, qlen - sch->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
696 del_timer(&q->perturb_timer);
697 if (q->perturb_period) {
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700698 mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
Stephen Hemmingerd46f8dd2008-01-20 17:19:43 -0800699 q->perturbation = net_random();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
701 sch_tree_unlock(sch);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000702 kfree(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 return 0;
704}
705
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000706static void *sfq_alloc(size_t sz)
707{
708 void *ptr = kmalloc(sz, GFP_KERNEL | __GFP_NOWARN);
709
710 if (!ptr)
711 ptr = vmalloc(sz);
712 return ptr;
713}
714
715static void sfq_free(void *addr)
716{
717 if (addr) {
718 if (is_vmalloc_addr(addr))
719 vfree(addr);
720 else
721 kfree(addr);
722 }
723}
724
725static void sfq_destroy(struct Qdisc *sch)
726{
727 struct sfq_sched_data *q = qdisc_priv(sch);
728
729 tcf_destroy_chain(&q->filter_list);
730 q->perturb_period = 0;
731 del_timer_sync(&q->perturb_timer);
732 sfq_free(q->ht);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000733 sfq_free(q->slots);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000734 kfree(q->red_parms);
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000735}
736
Patrick McHardy1e904742008-01-22 22:11:17 -0800737static int sfq_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
739 struct sfq_sched_data *q = qdisc_priv(sch);
740 int i;
741
Stephen Hemmingerd3e99482008-01-20 17:18:45 -0800742 q->perturb_timer.function = sfq_perturbation;
Fernando Carrijoc19a28e2009-01-07 18:09:08 -0800743 q->perturb_timer.data = (unsigned long)sch;
Stephen Hemmingerd3e99482008-01-20 17:18:45 -0800744 init_timer_deferrable(&q->perturb_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
Eric Dumazet18cb8092012-01-04 14:18:38 +0000746 for (i = 0; i < SFQ_MAX_DEPTH + 1; i++) {
747 q->dep[i].next = i + SFQ_MAX_FLOWS;
748 q->dep[i].prev = i + SFQ_MAX_FLOWS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800750
Eric Dumazet18cb8092012-01-04 14:18:38 +0000751 q->limit = SFQ_MAX_DEPTH;
752 q->maxdepth = SFQ_MAX_DEPTH;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000753 q->cur_depth = 0;
754 q->tail = NULL;
Eric Dumazet817fb152011-01-20 00:14:58 +0000755 q->divisor = SFQ_DEFAULT_HASH_DIVISOR;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000756 q->maxflows = SFQ_DEFAULT_FLOWS;
Eric Dumazet02a90982012-01-04 06:23:01 +0000757 q->quantum = psched_mtu(qdisc_dev(sch));
758 q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum);
759 q->perturb_period = 0;
760 q->perturbation = net_random();
761
762 if (opt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 int err = sfq_change(sch, opt);
764 if (err)
765 return err;
766 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800767
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000768 q->ht = sfq_alloc(sizeof(q->ht[0]) * q->divisor);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000769 q->slots = sfq_alloc(sizeof(q->slots[0]) * q->maxflows);
770 if (!q->ht || !q->slots) {
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000771 sfq_destroy(sch);
Eric Dumazet817fb152011-01-20 00:14:58 +0000772 return -ENOMEM;
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000773 }
Eric Dumazet817fb152011-01-20 00:14:58 +0000774 for (i = 0; i < q->divisor; i++)
775 q->ht[i] = SFQ_EMPTY_SLOT;
776
Eric Dumazet18cb8092012-01-04 14:18:38 +0000777 for (i = 0; i < q->maxflows; i++) {
Eric Dumazet18c8d822010-12-31 12:48:55 -0800778 slot_queue_init(&q->slots[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 sfq_link(q, i);
Eric Dumazet18c8d822010-12-31 12:48:55 -0800780 }
Eric Dumazet23624932011-01-21 16:26:09 -0800781 if (q->limit >= 1)
782 sch->flags |= TCQ_F_CAN_BYPASS;
783 else
784 sch->flags &= ~TCQ_F_CAN_BYPASS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 return 0;
786}
787
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788static int sfq_dump(struct Qdisc *sch, struct sk_buff *skb)
789{
790 struct sfq_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700791 unsigned char *b = skb_tail_pointer(skb);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000792 struct tc_sfq_qopt_v1 opt;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000793 struct red_parms *p = q->red_parms;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
Eric Dumazet18cb8092012-01-04 14:18:38 +0000795 memset(&opt, 0, sizeof(opt));
796 opt.v0.quantum = q->quantum;
797 opt.v0.perturb_period = q->perturb_period / HZ;
798 opt.v0.limit = q->limit;
799 opt.v0.divisor = q->divisor;
800 opt.v0.flows = q->maxflows;
801 opt.depth = q->maxdepth;
802 opt.headdrop = q->headdrop;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000804 if (p) {
805 opt.qth_min = p->qth_min >> p->Wlog;
806 opt.qth_max = p->qth_max >> p->Wlog;
807 opt.Wlog = p->Wlog;
808 opt.Plog = p->Plog;
809 opt.Scell_log = p->Scell_log;
810 opt.max_P = p->max_P;
811 }
812 memcpy(&opt.stats, &q->stats, sizeof(opt.stats));
813 opt.flags = q->flags;
814
Patrick McHardy1e904742008-01-22 22:11:17 -0800815 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
817 return skb->len;
818
Patrick McHardy1e904742008-01-22 22:11:17 -0800819nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700820 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 return -1;
822}
823
Jarek Poplawski41065fb2010-08-10 22:31:02 +0000824static struct Qdisc *sfq_leaf(struct Qdisc *sch, unsigned long arg)
825{
826 return NULL;
827}
828
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800829static unsigned long sfq_get(struct Qdisc *sch, u32 classid)
830{
831 return 0;
832}
833
Jarek Poplawskieb4a5522010-08-06 00:22:35 +0000834static unsigned long sfq_bind(struct Qdisc *sch, unsigned long parent,
835 u32 classid)
836{
Eric Dumazet23624932011-01-21 16:26:09 -0800837 /* we cannot bypass queue discipline anymore */
838 sch->flags &= ~TCQ_F_CAN_BYPASS;
Jarek Poplawskieb4a5522010-08-06 00:22:35 +0000839 return 0;
840}
841
Jarek Poplawskida7115d2010-08-09 12:18:17 +0000842static void sfq_put(struct Qdisc *q, unsigned long cl)
843{
844}
845
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800846static struct tcf_proto **sfq_find_tcf(struct Qdisc *sch, unsigned long cl)
847{
848 struct sfq_sched_data *q = qdisc_priv(sch);
849
850 if (cl)
851 return NULL;
852 return &q->filter_list;
853}
854
Patrick McHardy94de78d2008-01-31 18:37:16 -0800855static int sfq_dump_class(struct Qdisc *sch, unsigned long cl,
856 struct sk_buff *skb, struct tcmsg *tcm)
857{
858 tcm->tcm_handle |= TC_H_MIN(cl);
859 return 0;
860}
861
862static int sfq_dump_class_stats(struct Qdisc *sch, unsigned long cl,
863 struct gnet_dump *d)
864{
865 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800866 sfq_index idx = q->ht[cl - 1];
867 struct gnet_stats_queue qs = { 0 };
868 struct tc_sfq_xstats xstats = { 0 };
Eric Dumazetc4266262010-12-15 08:18:36 +0000869
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800870 if (idx != SFQ_EMPTY_SLOT) {
871 const struct sfq_slot *slot = &q->slots[idx];
Patrick McHardy94de78d2008-01-31 18:37:16 -0800872
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000873 xstats.allot = slot->allot << SFQ_ALLOT_SHIFT;
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800874 qs.qlen = slot->qlen;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000875 qs.backlog = slot->backlog;
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800876 }
Patrick McHardy94de78d2008-01-31 18:37:16 -0800877 if (gnet_stats_copy_queue(d, &qs) < 0)
878 return -1;
879 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
880}
881
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800882static void sfq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
883{
Patrick McHardy94de78d2008-01-31 18:37:16 -0800884 struct sfq_sched_data *q = qdisc_priv(sch);
885 unsigned int i;
886
887 if (arg->stop)
888 return;
889
Eric Dumazet817fb152011-01-20 00:14:58 +0000890 for (i = 0; i < q->divisor; i++) {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000891 if (q->ht[i] == SFQ_EMPTY_SLOT ||
Patrick McHardy94de78d2008-01-31 18:37:16 -0800892 arg->count < arg->skip) {
893 arg->count++;
894 continue;
895 }
896 if (arg->fn(sch, i + 1, arg) < 0) {
897 arg->stop = 1;
898 break;
899 }
900 arg->count++;
901 }
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800902}
903
904static const struct Qdisc_class_ops sfq_class_ops = {
Jarek Poplawski41065fb2010-08-10 22:31:02 +0000905 .leaf = sfq_leaf,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800906 .get = sfq_get,
Jarek Poplawskida7115d2010-08-09 12:18:17 +0000907 .put = sfq_put,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800908 .tcf_chain = sfq_find_tcf,
Jarek Poplawskieb4a5522010-08-06 00:22:35 +0000909 .bind_tcf = sfq_bind,
Jarek Poplawskida7115d2010-08-09 12:18:17 +0000910 .unbind_tcf = sfq_put,
Patrick McHardy94de78d2008-01-31 18:37:16 -0800911 .dump = sfq_dump_class,
912 .dump_stats = sfq_dump_class_stats,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800913 .walk = sfq_walk,
914};
915
Eric Dumazet20fea082007-11-14 01:44:41 -0800916static struct Qdisc_ops sfq_qdisc_ops __read_mostly = {
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800917 .cl_ops = &sfq_class_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 .id = "sfq",
919 .priv_size = sizeof(struct sfq_sched_data),
920 .enqueue = sfq_enqueue,
921 .dequeue = sfq_dequeue,
Eric Dumazet07bd8df2011-05-25 04:40:11 +0000922 .peek = qdisc_peek_dequeued,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 .drop = sfq_drop,
924 .init = sfq_init,
925 .reset = sfq_reset,
926 .destroy = sfq_destroy,
927 .change = NULL,
928 .dump = sfq_dump,
929 .owner = THIS_MODULE,
930};
931
932static int __init sfq_module_init(void)
933{
934 return register_qdisc(&sfq_qdisc_ops);
935}
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900936static void __exit sfq_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
938 unregister_qdisc(&sfq_qdisc_ops);
939}
940module_init(sfq_module_init)
941module_exit(sfq_module_exit)
942MODULE_LICENSE("GPL");