blob: 1af2f73906d07aa5bf6f428ffe37b81523178ee4 [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) \
Yang Yingliangfa089432013-12-10 20:55:33 +0800240 do { \
241 n = q->slots[x].dep.next; \
242 p = q->slots[x].dep.prev; \
243 sfq_dep_head(q, p)->next = n; \
244 sfq_dep_head(q, n)->prev = p; \
245 } while (0)
Eric Dumazeteda83e32010-12-20 12:54:58 +0000246
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248static inline void sfq_dec(struct sfq_sched_data *q, sfq_index x)
249{
250 sfq_index p, n;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000251 int d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
Eric Dumazeteda83e32010-12-20 12:54:58 +0000253 sfq_unlink(q, x, n, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
Eric Dumazeteda83e32010-12-20 12:54:58 +0000255 d = q->slots[x].qlen--;
256 if (n == p && q->cur_depth == d)
257 q->cur_depth--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 sfq_link(q, x);
259}
260
261static inline void sfq_inc(struct sfq_sched_data *q, sfq_index x)
262{
263 sfq_index p, n;
264 int d;
265
Eric Dumazeteda83e32010-12-20 12:54:58 +0000266 sfq_unlink(q, x, n, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Eric Dumazeteda83e32010-12-20 12:54:58 +0000268 d = ++q->slots[x].qlen;
269 if (q->cur_depth < d)
270 q->cur_depth = d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 sfq_link(q, x);
272}
273
Eric Dumazeteda83e32010-12-20 12:54:58 +0000274/* helper functions : might be changed when/if skb use a standard list_head */
275
276/* remove one skb from tail of slot queue */
277static inline struct sk_buff *slot_dequeue_tail(struct sfq_slot *slot)
278{
279 struct sk_buff *skb = slot->skblist_prev;
280
281 slot->skblist_prev = skb->prev;
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800282 skb->prev->next = (struct sk_buff *)slot;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000283 skb->next = skb->prev = NULL;
284 return skb;
285}
286
287/* remove one skb from head of slot queue */
288static inline struct sk_buff *slot_dequeue_head(struct sfq_slot *slot)
289{
290 struct sk_buff *skb = slot->skblist_next;
291
292 slot->skblist_next = skb->next;
Eric Dumazet18c8d822010-12-31 12:48:55 -0800293 skb->next->prev = (struct sk_buff *)slot;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000294 skb->next = skb->prev = NULL;
295 return skb;
296}
297
298static inline void slot_queue_init(struct sfq_slot *slot)
299{
Eric Dumazet18cb8092012-01-04 14:18:38 +0000300 memset(slot, 0, sizeof(*slot));
Eric Dumazeteda83e32010-12-20 12:54:58 +0000301 slot->skblist_prev = slot->skblist_next = (struct sk_buff *)slot;
302}
303
304/* add skb to slot queue (tail add) */
305static inline void slot_queue_add(struct sfq_slot *slot, struct sk_buff *skb)
306{
307 skb->prev = slot->skblist_prev;
308 skb->next = (struct sk_buff *)slot;
309 slot->skblist_prev->next = skb;
310 slot->skblist_prev = skb;
311}
312
313#define slot_queue_walk(slot, skb) \
314 for (skb = slot->skblist_next; \
315 skb != (struct sk_buff *)slot; \
316 skb = skb->next)
317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318static unsigned int sfq_drop(struct Qdisc *sch)
319{
320 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazeteda83e32010-12-20 12:54:58 +0000321 sfq_index x, d = q->cur_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 struct sk_buff *skb;
323 unsigned int len;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000324 struct sfq_slot *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
Eric Dumazeteda83e32010-12-20 12:54:58 +0000326 /* Queue is full! Find the longest slot and drop tail packet from it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 if (d > 1) {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000328 x = q->dep[d].next;
329 slot = &q->slots[x];
330drop:
Eric Dumazet18cb8092012-01-04 14:18:38 +0000331 skb = q->headdrop ? slot_dequeue_head(slot) : slot_dequeue_tail(slot);
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700332 len = qdisc_pkt_len(skb);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000333 slot->backlog -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 sfq_dec(q, x);
Eric Dumazeteda83e32010-12-20 12:54:58 +0000335 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 sch->q.qlen--;
337 sch->qstats.drops++;
Patrick McHardyf5539eb2006-03-20 19:01:38 -0800338 sch->qstats.backlog -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 return len;
340 }
341
342 if (d == 1) {
343 /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
Eric Dumazeteda83e32010-12-20 12:54:58 +0000344 x = q->tail->next;
345 slot = &q->slots[x];
346 q->tail->next = slot->next;
347 q->ht[slot->hash] = SFQ_EMPTY_SLOT;
348 goto drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 }
350
351 return 0;
352}
353
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000354/* Is ECN parameter configured */
355static int sfq_prob_mark(const struct sfq_sched_data *q)
356{
357 return q->flags & TC_RED_ECN;
358}
359
360/* Should packets over max threshold just be marked */
361static int sfq_hard_mark(const struct sfq_sched_data *q)
362{
363 return (q->flags & (TC_RED_ECN | TC_RED_HARDDROP)) == TC_RED_ECN;
364}
365
366static int sfq_headdrop(const struct sfq_sched_data *q)
367{
368 return q->headdrop;
369}
370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371static int
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800372sfq_enqueue(struct sk_buff *skb, struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
374 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800375 unsigned int hash;
Eric Dumazet8efa8852011-05-23 11:02:42 +0000376 sfq_index x, qlen;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000377 struct sfq_slot *slot;
Jarek Poplawski7f3ff4f2008-12-21 20:14:48 -0800378 int uninitialized_var(ret);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000379 struct sk_buff *head;
380 int delta;
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800381
382 hash = sfq_classify(skb, sch, &ret);
383 if (hash == 0) {
Jarek Poplawskic27f3392008-08-04 22:39:11 -0700384 if (ret & __NET_XMIT_BYPASS)
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800385 sch->qstats.drops++;
386 kfree_skb(skb);
387 return ret;
388 }
389 hash--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
391 x = q->ht[hash];
Eric Dumazeteda83e32010-12-20 12:54:58 +0000392 slot = &q->slots[x];
393 if (x == SFQ_EMPTY_SLOT) {
394 x = q->dep[0].next; /* get a free slot */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000395 if (x >= SFQ_MAX_FLOWS)
396 return qdisc_drop(skb, sch);
Eric Dumazeteda83e32010-12-20 12:54:58 +0000397 q->ht[hash] = x;
398 slot = &q->slots[x];
399 slot->hash = hash;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000400 slot->backlog = 0; /* should already be 0 anyway... */
401 red_set_vars(&slot->vars);
402 goto enqueue;
403 }
404 if (q->red_parms) {
405 slot->vars.qavg = red_calc_qavg_no_idle_time(q->red_parms,
406 &slot->vars,
407 slot->backlog);
408 switch (red_action(q->red_parms,
409 &slot->vars,
410 slot->vars.qavg)) {
411 case RED_DONT_MARK:
412 break;
413
414 case RED_PROB_MARK:
415 sch->qstats.overlimits++;
416 if (sfq_prob_mark(q)) {
417 /* We know we have at least one packet in queue */
418 if (sfq_headdrop(q) &&
419 INET_ECN_set_ce(slot->skblist_next)) {
420 q->stats.prob_mark_head++;
421 break;
422 }
423 if (INET_ECN_set_ce(skb)) {
424 q->stats.prob_mark++;
425 break;
426 }
427 }
428 q->stats.prob_drop++;
429 goto congestion_drop;
430
431 case RED_HARD_MARK:
432 sch->qstats.overlimits++;
433 if (sfq_hard_mark(q)) {
434 /* We know we have at least one packet in queue */
435 if (sfq_headdrop(q) &&
436 INET_ECN_set_ce(slot->skblist_next)) {
437 q->stats.forced_mark_head++;
438 break;
439 }
440 if (INET_ECN_set_ce(skb)) {
441 q->stats.forced_mark++;
442 break;
443 }
444 }
445 q->stats.forced_drop++;
446 goto congestion_drop;
447 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800449
Eric Dumazet18cb8092012-01-04 14:18:38 +0000450 if (slot->qlen >= q->maxdepth) {
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000451congestion_drop:
452 if (!sfq_headdrop(q))
Eric Dumazet18cb8092012-01-04 14:18:38 +0000453 return qdisc_drop(skb, sch);
454
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000455 /* We know we have at least one packet in queue */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000456 head = slot_dequeue_head(slot);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000457 delta = qdisc_pkt_len(head) - qdisc_pkt_len(skb);
458 sch->qstats.backlog -= delta;
459 slot->backlog -= delta;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000460 qdisc_drop(head, sch);
461
Eric Dumazet18cb8092012-01-04 14:18:38 +0000462 slot_queue_add(slot, skb);
463 return NET_XMIT_CN;
464 }
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700465
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000466enqueue:
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700467 sch->qstats.backlog += qdisc_pkt_len(skb);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000468 slot->backlog += qdisc_pkt_len(skb);
Eric Dumazeteda83e32010-12-20 12:54:58 +0000469 slot_queue_add(slot, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 sfq_inc(q, x);
Eric Dumazeteda83e32010-12-20 12:54:58 +0000471 if (slot->qlen == 1) { /* The flow is new */
472 if (q->tail == NULL) { /* It is the first flow */
473 slot->next = x;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 } else {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000475 slot->next = q->tail->next;
476 q->tail->next = x;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 }
Eric Dumazetcc34eb62012-03-13 18:04:25 +0000478 /* We put this flow at the end of our flow list.
479 * This might sound unfair for a new flow to wait after old ones,
480 * but we could endup servicing new flows only, and freeze old ones.
481 */
482 q->tail = slot;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000483 /* We could use a bigger initial quantum for new flows */
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000484 slot->allot = q->scaled_quantum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 }
Eric Dumazet9190b3b2011-01-20 23:31:33 -0800486 if (++sch->q.qlen <= q->limit)
Ben Greear9871e502010-08-10 01:45:40 -0700487 return NET_XMIT_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
Eric Dumazet8efa8852011-05-23 11:02:42 +0000489 qlen = slot->qlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 sfq_drop(sch);
Eric Dumazet8efa8852011-05-23 11:02:42 +0000491 /* Return Congestion Notification only if we dropped a packet
492 * from this flow.
493 */
Eric Dumazete1738bd2011-07-29 19:22:42 +0000494 if (qlen != slot->qlen)
495 return NET_XMIT_CN;
496
497 /* As we dropped a packet, better let upper stack know this */
498 qdisc_tree_decrease_qlen(sch, 1);
499 return NET_XMIT_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500}
501
Patrick McHardy48a8f512008-10-31 00:44:18 -0700502static struct sk_buff *
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800503sfq_dequeue(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
505 struct sfq_sched_data *q = qdisc_priv(sch);
506 struct sk_buff *skb;
Eric Dumazetaa3e2192010-12-20 13:18:16 -0800507 sfq_index a, next_a;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000508 struct sfq_slot *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510 /* No active slots */
Eric Dumazeteda83e32010-12-20 12:54:58 +0000511 if (q->tail == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return NULL;
513
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000514next_slot:
Eric Dumazeteda83e32010-12-20 12:54:58 +0000515 a = q->tail->next;
516 slot = &q->slots[a];
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000517 if (slot->allot <= 0) {
518 q->tail = slot;
519 slot->allot += q->scaled_quantum;
520 goto next_slot;
521 }
Eric Dumazeteda83e32010-12-20 12:54:58 +0000522 skb = slot_dequeue_head(slot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 sfq_dec(q, a);
Eric Dumazet9190b3b2011-01-20 23:31:33 -0800524 qdisc_bstats_update(sch, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 sch->q.qlen--;
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700526 sch->qstats.backlog -= qdisc_pkt_len(skb);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000527 slot->backlog -= qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 /* Is the slot empty? */
Eric Dumazeteda83e32010-12-20 12:54:58 +0000529 if (slot->qlen == 0) {
530 q->ht[slot->hash] = SFQ_EMPTY_SLOT;
531 next_a = slot->next;
Eric Dumazetaa3e2192010-12-20 13:18:16 -0800532 if (a == next_a) {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000533 q->tail = NULL; /* no more active slots */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 return skb;
535 }
Eric Dumazeteda83e32010-12-20 12:54:58 +0000536 q->tail->next = next_a;
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000537 } else {
538 slot->allot -= SFQ_ALLOT_SIZE(qdisc_pkt_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
540 return skb;
541}
542
543static void
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800544sfq_reset(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545{
546 struct sk_buff *skb;
547
548 while ((skb = sfq_dequeue(sch)) != NULL)
549 kfree_skb(skb);
550}
551
Eric Dumazet225d9b82011-12-21 03:30:11 +0000552/*
553 * When q->perturbation is changed, we rehash all queued skbs
554 * to avoid OOO (Out Of Order) effects.
555 * We dont use sfq_dequeue()/sfq_enqueue() because we dont want to change
556 * counters.
557 */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000558static void sfq_rehash(struct Qdisc *sch)
Eric Dumazet225d9b82011-12-21 03:30:11 +0000559{
Eric Dumazet18cb8092012-01-04 14:18:38 +0000560 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000561 struct sk_buff *skb;
562 int i;
563 struct sfq_slot *slot;
564 struct sk_buff_head list;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000565 int dropped = 0;
Eric Dumazet225d9b82011-12-21 03:30:11 +0000566
567 __skb_queue_head_init(&list);
568
Eric Dumazet18cb8092012-01-04 14:18:38 +0000569 for (i = 0; i < q->maxflows; i++) {
Eric Dumazet225d9b82011-12-21 03:30:11 +0000570 slot = &q->slots[i];
571 if (!slot->qlen)
572 continue;
573 while (slot->qlen) {
574 skb = slot_dequeue_head(slot);
575 sfq_dec(q, i);
576 __skb_queue_tail(&list, skb);
577 }
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000578 slot->backlog = 0;
579 red_set_vars(&slot->vars);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000580 q->ht[slot->hash] = SFQ_EMPTY_SLOT;
581 }
582 q->tail = NULL;
583
584 while ((skb = __skb_dequeue(&list)) != NULL) {
585 unsigned int hash = sfq_hash(q, skb);
586 sfq_index x = q->ht[hash];
587
588 slot = &q->slots[x];
589 if (x == SFQ_EMPTY_SLOT) {
590 x = q->dep[0].next; /* get a free slot */
Eric Dumazet18cb8092012-01-04 14:18:38 +0000591 if (x >= SFQ_MAX_FLOWS) {
592drop: sch->qstats.backlog -= qdisc_pkt_len(skb);
593 kfree_skb(skb);
594 dropped++;
595 continue;
596 }
Eric Dumazet225d9b82011-12-21 03:30:11 +0000597 q->ht[hash] = x;
598 slot = &q->slots[x];
599 slot->hash = hash;
600 }
Eric Dumazet18cb8092012-01-04 14:18:38 +0000601 if (slot->qlen >= q->maxdepth)
602 goto drop;
Eric Dumazet225d9b82011-12-21 03:30:11 +0000603 slot_queue_add(slot, skb);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000604 if (q->red_parms)
605 slot->vars.qavg = red_calc_qavg(q->red_parms,
606 &slot->vars,
607 slot->backlog);
608 slot->backlog += qdisc_pkt_len(skb);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000609 sfq_inc(q, x);
610 if (slot->qlen == 1) { /* The flow is new */
611 if (q->tail == NULL) { /* It is the first flow */
612 slot->next = x;
613 } else {
614 slot->next = q->tail->next;
615 q->tail->next = x;
616 }
617 q->tail = slot;
618 slot->allot = q->scaled_quantum;
619 }
620 }
Eric Dumazet18cb8092012-01-04 14:18:38 +0000621 sch->q.qlen -= dropped;
622 qdisc_tree_decrease_qlen(sch, dropped);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000623}
624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625static void sfq_perturbation(unsigned long arg)
626{
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800627 struct Qdisc *sch = (struct Qdisc *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000629 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
Eric Dumazet225d9b82011-12-21 03:30:11 +0000631 spin_lock(root_lock);
Aruna-Hewapathirane63862b52014-01-11 07:15:59 -0500632 q->perturbation = prandom_u32();
Eric Dumazet225d9b82011-12-21 03:30:11 +0000633 if (!q->filter_list && q->tail)
Eric Dumazet18cb8092012-01-04 14:18:38 +0000634 sfq_rehash(sch);
Eric Dumazet225d9b82011-12-21 03:30:11 +0000635 spin_unlock(root_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700637 if (q->perturb_period)
638 mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639}
640
Patrick McHardy1e904742008-01-22 22:11:17 -0800641static int sfq_change(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
643 struct sfq_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800644 struct tc_sfq_qopt *ctl = nla_data(opt);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000645 struct tc_sfq_qopt_v1 *ctl_v1 = NULL;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800646 unsigned int qlen;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000647 struct red_parms *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Patrick McHardy1e904742008-01-22 22:11:17 -0800649 if (opt->nla_len < nla_attr_size(sizeof(*ctl)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 return -EINVAL;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000651 if (opt->nla_len >= nla_attr_size(sizeof(*ctl_v1)))
652 ctl_v1 = nla_data(opt);
stephen hemminger119b3d32011-02-02 15:19:51 +0000653 if (ctl->divisor &&
654 (!is_power_of_2(ctl->divisor) || ctl->divisor > 65536))
655 return -EINVAL;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000656 if (ctl_v1 && ctl_v1->qth_min) {
657 p = kmalloc(sizeof(*p), GFP_KERNEL);
658 if (!p)
659 return -ENOMEM;
660 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 sch_tree_lock(sch);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000662 if (ctl->quantum) {
663 q->quantum = ctl->quantum;
664 q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum);
665 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800666 q->perturb_period = ctl->perturb_period * HZ;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000667 if (ctl->flows)
668 q->maxflows = min_t(u32, ctl->flows, SFQ_MAX_FLOWS);
669 if (ctl->divisor) {
Eric Dumazet817fb152011-01-20 00:14:58 +0000670 q->divisor = ctl->divisor;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000671 q->maxflows = min_t(u32, q->maxflows, q->divisor);
672 }
673 if (ctl_v1) {
674 if (ctl_v1->depth)
675 q->maxdepth = min_t(u32, ctl_v1->depth, SFQ_MAX_DEPTH);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000676 if (p) {
677 swap(q->red_parms, p);
678 red_set_parms(q->red_parms,
679 ctl_v1->qth_min, ctl_v1->qth_max,
680 ctl_v1->Wlog,
681 ctl_v1->Plog, ctl_v1->Scell_log,
682 NULL,
683 ctl_v1->max_P);
684 }
685 q->flags = ctl_v1->flags;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000686 q->headdrop = ctl_v1->headdrop;
687 }
688 if (ctl->limit) {
689 q->limit = min_t(u32, ctl->limit, q->maxdepth * q->maxflows);
690 q->maxflows = min_t(u32, q->maxflows, q->limit);
691 }
692
Patrick McHardy5e50da02006-11-29 17:36:20 -0800693 qlen = sch->q.qlen;
Alexey Kuznetsov5588b402007-09-19 10:42:03 -0700694 while (sch->q.qlen > q->limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 sfq_drop(sch);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800696 qdisc_tree_decrease_qlen(sch, qlen - sch->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698 del_timer(&q->perturb_timer);
699 if (q->perturb_period) {
Alexey Kuznetsov32740dd2007-09-30 17:51:33 -0700700 mod_timer(&q->perturb_timer, jiffies + q->perturb_period);
Aruna-Hewapathirane63862b52014-01-11 07:15:59 -0500701 q->perturbation = prandom_u32();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 }
703 sch_tree_unlock(sch);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000704 kfree(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 return 0;
706}
707
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000708static void *sfq_alloc(size_t sz)
709{
710 void *ptr = kmalloc(sz, GFP_KERNEL | __GFP_NOWARN);
711
712 if (!ptr)
713 ptr = vmalloc(sz);
714 return ptr;
715}
716
717static void sfq_free(void *addr)
718{
WANG Cong4cb28972014-06-02 15:55:22 -0700719 kvfree(addr);
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000720}
721
722static void sfq_destroy(struct Qdisc *sch)
723{
724 struct sfq_sched_data *q = qdisc_priv(sch);
725
726 tcf_destroy_chain(&q->filter_list);
727 q->perturb_period = 0;
728 del_timer_sync(&q->perturb_timer);
729 sfq_free(q->ht);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000730 sfq_free(q->slots);
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000731 kfree(q->red_parms);
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000732}
733
Patrick McHardy1e904742008-01-22 22:11:17 -0800734static int sfq_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
736 struct sfq_sched_data *q = qdisc_priv(sch);
737 int i;
738
Stephen Hemmingerd3e99482008-01-20 17:18:45 -0800739 q->perturb_timer.function = sfq_perturbation;
Fernando Carrijoc19a28e2009-01-07 18:09:08 -0800740 q->perturb_timer.data = (unsigned long)sch;
Stephen Hemmingerd3e99482008-01-20 17:18:45 -0800741 init_timer_deferrable(&q->perturb_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742
Eric Dumazet18cb8092012-01-04 14:18:38 +0000743 for (i = 0; i < SFQ_MAX_DEPTH + 1; i++) {
744 q->dep[i].next = i + SFQ_MAX_FLOWS;
745 q->dep[i].prev = i + SFQ_MAX_FLOWS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800747
Eric Dumazet18cb8092012-01-04 14:18:38 +0000748 q->limit = SFQ_MAX_DEPTH;
749 q->maxdepth = SFQ_MAX_DEPTH;
Eric Dumazeteda83e32010-12-20 12:54:58 +0000750 q->cur_depth = 0;
751 q->tail = NULL;
Eric Dumazet817fb152011-01-20 00:14:58 +0000752 q->divisor = SFQ_DEFAULT_HASH_DIVISOR;
Eric Dumazet18cb8092012-01-04 14:18:38 +0000753 q->maxflows = SFQ_DEFAULT_FLOWS;
Eric Dumazet02a90982012-01-04 06:23:01 +0000754 q->quantum = psched_mtu(qdisc_dev(sch));
755 q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum);
756 q->perturb_period = 0;
Aruna-Hewapathirane63862b52014-01-11 07:15:59 -0500757 q->perturbation = prandom_u32();
Eric Dumazet02a90982012-01-04 06:23:01 +0000758
759 if (opt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 int err = sfq_change(sch, opt);
761 if (err)
762 return err;
763 }
Stephen Hemminger6f9e98f2008-01-20 17:20:56 -0800764
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000765 q->ht = sfq_alloc(sizeof(q->ht[0]) * q->divisor);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000766 q->slots = sfq_alloc(sizeof(q->slots[0]) * q->maxflows);
767 if (!q->ht || !q->slots) {
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000768 sfq_destroy(sch);
Eric Dumazet817fb152011-01-20 00:14:58 +0000769 return -ENOMEM;
Eric Dumazetbd16a6c2012-01-04 06:22:24 +0000770 }
Eric Dumazet817fb152011-01-20 00:14:58 +0000771 for (i = 0; i < q->divisor; i++)
772 q->ht[i] = SFQ_EMPTY_SLOT;
773
Eric Dumazet18cb8092012-01-04 14:18:38 +0000774 for (i = 0; i < q->maxflows; i++) {
Eric Dumazet18c8d822010-12-31 12:48:55 -0800775 slot_queue_init(&q->slots[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 sfq_link(q, i);
Eric Dumazet18c8d822010-12-31 12:48:55 -0800777 }
Eric Dumazet23624932011-01-21 16:26:09 -0800778 if (q->limit >= 1)
779 sch->flags |= TCQ_F_CAN_BYPASS;
780 else
781 sch->flags &= ~TCQ_F_CAN_BYPASS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 return 0;
783}
784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785static int sfq_dump(struct Qdisc *sch, struct sk_buff *skb)
786{
787 struct sfq_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700788 unsigned char *b = skb_tail_pointer(skb);
Eric Dumazet18cb8092012-01-04 14:18:38 +0000789 struct tc_sfq_qopt_v1 opt;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000790 struct red_parms *p = q->red_parms;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
Eric Dumazet18cb8092012-01-04 14:18:38 +0000792 memset(&opt, 0, sizeof(opt));
793 opt.v0.quantum = q->quantum;
794 opt.v0.perturb_period = q->perturb_period / HZ;
795 opt.v0.limit = q->limit;
796 opt.v0.divisor = q->divisor;
797 opt.v0.flows = q->maxflows;
798 opt.depth = q->maxdepth;
799 opt.headdrop = q->headdrop;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000801 if (p) {
802 opt.qth_min = p->qth_min >> p->Wlog;
803 opt.qth_max = p->qth_max >> p->Wlog;
804 opt.Wlog = p->Wlog;
805 opt.Plog = p->Plog;
806 opt.Scell_log = p->Scell_log;
807 opt.max_P = p->max_P;
808 }
809 memcpy(&opt.stats, &q->stats, sizeof(opt.stats));
810 opt.flags = q->flags;
811
David S. Miller1b34ec42012-03-29 05:11:39 -0400812 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
813 goto nla_put_failure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
815 return skb->len;
816
Patrick McHardy1e904742008-01-22 22:11:17 -0800817nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700818 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 return -1;
820}
821
Jarek Poplawski41065fb2010-08-10 22:31:02 +0000822static struct Qdisc *sfq_leaf(struct Qdisc *sch, unsigned long arg)
823{
824 return NULL;
825}
826
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800827static unsigned long sfq_get(struct Qdisc *sch, u32 classid)
828{
829 return 0;
830}
831
Jarek Poplawskieb4a5522010-08-06 00:22:35 +0000832static unsigned long sfq_bind(struct Qdisc *sch, unsigned long parent,
833 u32 classid)
834{
Eric Dumazet23624932011-01-21 16:26:09 -0800835 /* we cannot bypass queue discipline anymore */
836 sch->flags &= ~TCQ_F_CAN_BYPASS;
Jarek Poplawskieb4a5522010-08-06 00:22:35 +0000837 return 0;
838}
839
Jarek Poplawskida7115d2010-08-09 12:18:17 +0000840static void sfq_put(struct Qdisc *q, unsigned long cl)
841{
842}
843
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800844static struct tcf_proto **sfq_find_tcf(struct Qdisc *sch, unsigned long cl)
845{
846 struct sfq_sched_data *q = qdisc_priv(sch);
847
848 if (cl)
849 return NULL;
850 return &q->filter_list;
851}
852
Patrick McHardy94de78d2008-01-31 18:37:16 -0800853static int sfq_dump_class(struct Qdisc *sch, unsigned long cl,
854 struct sk_buff *skb, struct tcmsg *tcm)
855{
856 tcm->tcm_handle |= TC_H_MIN(cl);
857 return 0;
858}
859
860static int sfq_dump_class_stats(struct Qdisc *sch, unsigned long cl,
861 struct gnet_dump *d)
862{
863 struct sfq_sched_data *q = qdisc_priv(sch);
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800864 sfq_index idx = q->ht[cl - 1];
865 struct gnet_stats_queue qs = { 0 };
866 struct tc_sfq_xstats xstats = { 0 };
Eric Dumazetc4266262010-12-15 08:18:36 +0000867
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800868 if (idx != SFQ_EMPTY_SLOT) {
869 const struct sfq_slot *slot = &q->slots[idx];
Patrick McHardy94de78d2008-01-31 18:37:16 -0800870
Eric Dumazeteeaeb062010-12-28 21:53:33 +0000871 xstats.allot = slot->allot << SFQ_ALLOT_SHIFT;
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800872 qs.qlen = slot->qlen;
Eric Dumazetddecf0f2012-01-06 06:31:44 +0000873 qs.backlog = slot->backlog;
Eric Dumazetee09b3c2010-12-22 11:39:59 -0800874 }
Patrick McHardy94de78d2008-01-31 18:37:16 -0800875 if (gnet_stats_copy_queue(d, &qs) < 0)
876 return -1;
877 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
878}
879
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800880static void sfq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
881{
Patrick McHardy94de78d2008-01-31 18:37:16 -0800882 struct sfq_sched_data *q = qdisc_priv(sch);
883 unsigned int i;
884
885 if (arg->stop)
886 return;
887
Eric Dumazet817fb152011-01-20 00:14:58 +0000888 for (i = 0; i < q->divisor; i++) {
Eric Dumazeteda83e32010-12-20 12:54:58 +0000889 if (q->ht[i] == SFQ_EMPTY_SLOT ||
Patrick McHardy94de78d2008-01-31 18:37:16 -0800890 arg->count < arg->skip) {
891 arg->count++;
892 continue;
893 }
894 if (arg->fn(sch, i + 1, arg) < 0) {
895 arg->stop = 1;
896 break;
897 }
898 arg->count++;
899 }
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800900}
901
902static const struct Qdisc_class_ops sfq_class_ops = {
Jarek Poplawski41065fb2010-08-10 22:31:02 +0000903 .leaf = sfq_leaf,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800904 .get = sfq_get,
Jarek Poplawskida7115d2010-08-09 12:18:17 +0000905 .put = sfq_put,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800906 .tcf_chain = sfq_find_tcf,
Jarek Poplawskieb4a5522010-08-06 00:22:35 +0000907 .bind_tcf = sfq_bind,
Jarek Poplawskida7115d2010-08-09 12:18:17 +0000908 .unbind_tcf = sfq_put,
Patrick McHardy94de78d2008-01-31 18:37:16 -0800909 .dump = sfq_dump_class,
910 .dump_stats = sfq_dump_class_stats,
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800911 .walk = sfq_walk,
912};
913
Eric Dumazet20fea082007-11-14 01:44:41 -0800914static struct Qdisc_ops sfq_qdisc_ops __read_mostly = {
Patrick McHardy7d2681a2008-01-31 18:36:52 -0800915 .cl_ops = &sfq_class_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 .id = "sfq",
917 .priv_size = sizeof(struct sfq_sched_data),
918 .enqueue = sfq_enqueue,
919 .dequeue = sfq_dequeue,
Eric Dumazet07bd8df2011-05-25 04:40:11 +0000920 .peek = qdisc_peek_dequeued,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 .drop = sfq_drop,
922 .init = sfq_init,
923 .reset = sfq_reset,
924 .destroy = sfq_destroy,
925 .change = NULL,
926 .dump = sfq_dump,
927 .owner = THIS_MODULE,
928};
929
930static int __init sfq_module_init(void)
931{
932 return register_qdisc(&sfq_qdisc_ops);
933}
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900934static void __exit sfq_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935{
936 unregister_qdisc(&sfq_qdisc_ops);
937}
938module_init(sfq_module_init)
939module_exit(sfq_module_exit)
940MODULE_LICENSE("GPL");