blob: f377702d4b9185762293a7251f2574a9e72515eb [file] [log] [blame]
Eric Dumazetafe4fd02013-08-29 15:49:55 -07001/*
2 * net/sched/sch_fq.c Fair Queue Packet Scheduler (per flow pacing)
3 *
Eric Dumazet86b3bfe2015-01-28 06:06:36 -08004 * Copyright (C) 2013-2015 Eric Dumazet <edumazet@google.com>
Eric Dumazetafe4fd02013-08-29 15:49:55 -07005 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
Simon Horman05e8bb82015-04-02 11:20:23 +090011 * Meant to be mostly used for locally generated traffic :
Eric Dumazetafe4fd02013-08-29 15:49:55 -070012 * Fast classification depends on skb->sk being set before reaching us.
13 * If not, (router workload), we use rxhash as fallback, with 32 bits wide hash.
14 * All packets belonging to a socket are considered as a 'flow'.
15 *
16 * Flows are dynamically allocated and stored in a hash table of RB trees
17 * They are also part of one Round Robin 'queues' (new or old flows)
18 *
19 * Burst avoidance (aka pacing) capability :
20 *
21 * Transport (eg TCP) can set in sk->sk_pacing_rate a rate, enqueue a
22 * bunch of packets, and this packet scheduler adds delay between
23 * packets to respect rate limitation.
24 *
25 * enqueue() :
26 * - lookup one RB tree (out of 1024 or more) to find the flow.
27 * If non existent flow, create it, add it to the tree.
28 * Add skb to the per flow list of skb (fifo).
29 * - Use a special fifo for high prio packets
30 *
31 * dequeue() : serves flows in Round Robin
32 * Note : When a flow becomes empty, we do not immediately remove it from
33 * rb trees, for performance reasons (its expected to send additional packets,
34 * or SLAB cache will reuse socket for another flow)
35 */
36
37#include <linux/module.h>
38#include <linux/types.h>
39#include <linux/kernel.h>
40#include <linux/jiffies.h>
41#include <linux/string.h>
42#include <linux/in.h>
43#include <linux/errno.h>
44#include <linux/init.h>
45#include <linux/skbuff.h>
46#include <linux/slab.h>
47#include <linux/rbtree.h>
48#include <linux/hash.h>
Eric Dumazet08f89b92013-08-30 09:46:43 -070049#include <linux/prefetch.h>
Eric Dumazetc3bd8542013-12-15 13:15:25 -080050#include <linux/vmalloc.h>
Eric Dumazetafe4fd02013-08-29 15:49:55 -070051#include <net/netlink.h>
52#include <net/pkt_sched.h>
53#include <net/sock.h>
54#include <net/tcp_states.h>
Eric Dumazet98781962015-02-03 18:31:53 -080055#include <net/tcp.h>
Eric Dumazetafe4fd02013-08-29 15:49:55 -070056
57/*
58 * Per flow structure, dynamically allocated
59 */
60struct fq_flow {
61 struct sk_buff *head; /* list of skbs for this flow : first skb */
62 union {
63 struct sk_buff *tail; /* last skb in the list */
64 unsigned long age; /* jiffies when flow was emptied, for gc */
65 };
Simon Horman05e8bb82015-04-02 11:20:23 +090066 struct rb_node fq_node; /* anchor in fq_root[] trees */
Eric Dumazetafe4fd02013-08-29 15:49:55 -070067 struct sock *sk;
68 int qlen; /* number of packets in flow queue */
69 int credit;
70 u32 socket_hash; /* sk_hash */
71 struct fq_flow *next; /* next pointer in RR lists, or &detached */
72
73 struct rb_node rate_node; /* anchor in q->delayed tree */
74 u64 time_next_packet;
75};
76
77struct fq_flow_head {
78 struct fq_flow *first;
79 struct fq_flow *last;
80};
81
82struct fq_sched_data {
83 struct fq_flow_head new_flows;
84
85 struct fq_flow_head old_flows;
86
87 struct rb_root delayed; /* for rate limited flows */
88 u64 time_next_delayed_flow;
89
90 struct fq_flow internal; /* for non classified or high prio packets */
91 u32 quantum;
92 u32 initial_quantum;
Eric Dumazetf52ed892013-11-15 08:58:14 -080093 u32 flow_refill_delay;
Eric Dumazetafe4fd02013-08-29 15:49:55 -070094 u32 flow_max_rate; /* optional max rate per flow */
95 u32 flow_plimit; /* max packets per flow */
Eric Dumazet06eb3952015-02-04 21:30:40 -080096 u32 orphan_mask; /* mask for orphaned skb */
Eric Dumazetafe4fd02013-08-29 15:49:55 -070097 struct rb_root *fq_root;
98 u8 rate_enable;
99 u8 fq_trees_log;
100
101 u32 flows;
102 u32 inactive_flows;
103 u32 throttled_flows;
104
105 u64 stat_gc_flows;
106 u64 stat_internal_packets;
107 u64 stat_tcp_retrans;
108 u64 stat_throttled;
109 u64 stat_flows_plimit;
110 u64 stat_pkts_too_long;
111 u64 stat_allocation_errors;
112 struct qdisc_watchdog watchdog;
113};
114
115/* special value to mark a detached flow (not on old/new list) */
116static struct fq_flow detached, throttled;
117
118static void fq_flow_set_detached(struct fq_flow *f)
119{
120 f->next = &detached;
Eric Dumazetf52ed892013-11-15 08:58:14 -0800121 f->age = jiffies;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700122}
123
124static bool fq_flow_is_detached(const struct fq_flow *f)
125{
126 return f->next == &detached;
127}
128
129static void fq_flow_set_throttled(struct fq_sched_data *q, struct fq_flow *f)
130{
131 struct rb_node **p = &q->delayed.rb_node, *parent = NULL;
132
133 while (*p) {
134 struct fq_flow *aux;
135
136 parent = *p;
137 aux = container_of(parent, struct fq_flow, rate_node);
138 if (f->time_next_packet >= aux->time_next_packet)
139 p = &parent->rb_right;
140 else
141 p = &parent->rb_left;
142 }
143 rb_link_node(&f->rate_node, parent, p);
144 rb_insert_color(&f->rate_node, &q->delayed);
145 q->throttled_flows++;
146 q->stat_throttled++;
147
148 f->next = &throttled;
149 if (q->time_next_delayed_flow > f->time_next_packet)
150 q->time_next_delayed_flow = f->time_next_packet;
151}
152
153
154static struct kmem_cache *fq_flow_cachep __read_mostly;
155
156static void fq_flow_add_tail(struct fq_flow_head *head, struct fq_flow *flow)
157{
158 if (head->first)
159 head->last->next = flow;
160 else
161 head->first = flow;
162 head->last = flow;
163 flow->next = NULL;
164}
165
166/* limit number of collected flows per round */
167#define FQ_GC_MAX 8
168#define FQ_GC_AGE (3*HZ)
169
170static bool fq_gc_candidate(const struct fq_flow *f)
171{
172 return fq_flow_is_detached(f) &&
173 time_after(jiffies, f->age + FQ_GC_AGE);
174}
175
176static void fq_gc(struct fq_sched_data *q,
177 struct rb_root *root,
178 struct sock *sk)
179{
180 struct fq_flow *f, *tofree[FQ_GC_MAX];
181 struct rb_node **p, *parent;
182 int fcnt = 0;
183
184 p = &root->rb_node;
185 parent = NULL;
186 while (*p) {
187 parent = *p;
188
189 f = container_of(parent, struct fq_flow, fq_node);
190 if (f->sk == sk)
191 break;
192
193 if (fq_gc_candidate(f)) {
194 tofree[fcnt++] = f;
195 if (fcnt == FQ_GC_MAX)
196 break;
197 }
198
199 if (f->sk > sk)
200 p = &parent->rb_right;
201 else
202 p = &parent->rb_left;
203 }
204
205 q->flows -= fcnt;
206 q->inactive_flows -= fcnt;
207 q->stat_gc_flows += fcnt;
208 while (fcnt) {
209 struct fq_flow *f = tofree[--fcnt];
210
211 rb_erase(&f->fq_node, root);
212 kmem_cache_free(fq_flow_cachep, f);
213 }
214}
215
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700216static struct fq_flow *fq_classify(struct sk_buff *skb, struct fq_sched_data *q)
217{
218 struct rb_node **p, *parent;
219 struct sock *sk = skb->sk;
220 struct rb_root *root;
221 struct fq_flow *f;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700222
223 /* warning: no starvation prevention... */
Maciej Żenczykowski2abc2f02013-11-14 08:50:43 -0800224 if (unlikely((skb->priority & TC_PRIO_MAX) == TC_PRIO_CONTROL))
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700225 return &q->internal;
226
Eric Dumazet06eb3952015-02-04 21:30:40 -0800227 /* SYNACK messages are attached to a listener socket.
228 * 1) They are not part of a 'flow' yet
229 * 2) We do not want to rate limit them (eg SYNFLOOD attack),
230 * especially if the listener set SO_MAX_PACING_RATE
231 * 3) We pretend they are orphaned
232 */
233 if (!sk || sk->sk_state == TCP_LISTEN) {
234 unsigned long hash = skb_get_hash(skb) & q->orphan_mask;
235
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700236 /* By forcing low order bit to 1, we make sure to not
237 * collide with a local flow (socket pointers are word aligned)
238 */
Eric Dumazet06eb3952015-02-04 21:30:40 -0800239 sk = (struct sock *)((hash << 1) | 1UL);
240 skb_orphan(skb);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700241 }
242
243 root = &q->fq_root[hash_32((u32)(long)sk, q->fq_trees_log)];
244
245 if (q->flows >= (2U << q->fq_trees_log) &&
246 q->inactive_flows > q->flows/2)
247 fq_gc(q, root, sk);
248
249 p = &root->rb_node;
250 parent = NULL;
251 while (*p) {
252 parent = *p;
253
254 f = container_of(parent, struct fq_flow, fq_node);
255 if (f->sk == sk) {
256 /* socket might have been reallocated, so check
257 * if its sk_hash is the same.
258 * It not, we need to refill credit with
259 * initial quantum
260 */
261 if (unlikely(skb->sk &&
262 f->socket_hash != sk->sk_hash)) {
263 f->credit = q->initial_quantum;
264 f->socket_hash = sk->sk_hash;
Eric Dumazetfc59d5b2013-10-27 16:26:43 -0700265 f->time_next_packet = 0ULL;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700266 }
267 return f;
268 }
269 if (f->sk > sk)
270 p = &parent->rb_right;
271 else
272 p = &parent->rb_left;
273 }
274
275 f = kmem_cache_zalloc(fq_flow_cachep, GFP_ATOMIC | __GFP_NOWARN);
276 if (unlikely(!f)) {
277 q->stat_allocation_errors++;
278 return &q->internal;
279 }
280 fq_flow_set_detached(f);
281 f->sk = sk;
282 if (skb->sk)
283 f->socket_hash = sk->sk_hash;
284 f->credit = q->initial_quantum;
285
286 rb_link_node(&f->fq_node, parent, p);
287 rb_insert_color(&f->fq_node, root);
288
289 q->flows++;
290 q->inactive_flows++;
291 return f;
292}
293
294
295/* remove one skb from head of flow queue */
Eric Dumazet8d34ce12013-09-27 14:20:01 -0700296static struct sk_buff *fq_dequeue_head(struct Qdisc *sch, struct fq_flow *flow)
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700297{
298 struct sk_buff *skb = flow->head;
299
300 if (skb) {
301 flow->head = skb->next;
302 skb->next = NULL;
303 flow->qlen--;
John Fastabend25331d62014-09-28 11:53:29 -0700304 qdisc_qstats_backlog_dec(sch, skb);
Eric Dumazet8d34ce12013-09-27 14:20:01 -0700305 sch->q.qlen--;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700306 }
307 return skb;
308}
309
310/* We might add in the future detection of retransmits
311 * For the time being, just return false
312 */
313static bool skb_is_retransmit(struct sk_buff *skb)
314{
315 return false;
316}
317
318/* add skb to flow queue
319 * flow queue is a linked list, kind of FIFO, except for TCP retransmits
320 * We special case tcp retransmits to be transmitted before other packets.
321 * We rely on fact that TCP retransmits are unlikely, so we do not waste
322 * a separate queue or a pointer.
323 * head-> [retrans pkt 1]
324 * [retrans pkt 2]
325 * [ normal pkt 1]
326 * [ normal pkt 2]
327 * [ normal pkt 3]
328 * tail-> [ normal pkt 4]
329 */
330static void flow_queue_add(struct fq_flow *flow, struct sk_buff *skb)
331{
332 struct sk_buff *prev, *head = flow->head;
333
334 skb->next = NULL;
335 if (!head) {
336 flow->head = skb;
337 flow->tail = skb;
338 return;
339 }
340 if (likely(!skb_is_retransmit(skb))) {
341 flow->tail->next = skb;
342 flow->tail = skb;
343 return;
344 }
345
346 /* This skb is a tcp retransmit,
347 * find the last retrans packet in the queue
348 */
349 prev = NULL;
350 while (skb_is_retransmit(head)) {
351 prev = head;
352 head = head->next;
353 if (!head)
354 break;
355 }
356 if (!prev) { /* no rtx packet in queue, become the new head */
357 skb->next = flow->head;
358 flow->head = skb;
359 } else {
360 if (prev == flow->tail)
361 flow->tail = skb;
362 else
363 skb->next = prev->next;
364 prev->next = skb;
365 }
366}
367
368static int fq_enqueue(struct sk_buff *skb, struct Qdisc *sch)
369{
370 struct fq_sched_data *q = qdisc_priv(sch);
371 struct fq_flow *f;
372
373 if (unlikely(sch->q.qlen >= sch->limit))
374 return qdisc_drop(skb, sch);
375
376 f = fq_classify(skb, q);
377 if (unlikely(f->qlen >= q->flow_plimit && f != &q->internal)) {
378 q->stat_flows_plimit++;
379 return qdisc_drop(skb, sch);
380 }
381
382 f->qlen++;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700383 if (skb_is_retransmit(skb))
384 q->stat_tcp_retrans++;
John Fastabend25331d62014-09-28 11:53:29 -0700385 qdisc_qstats_backlog_inc(sch, skb);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700386 if (fq_flow_is_detached(f)) {
387 fq_flow_add_tail(&q->new_flows, f);
Eric Dumazetf52ed892013-11-15 08:58:14 -0800388 if (time_after(jiffies, f->age + q->flow_refill_delay))
389 f->credit = max_t(u32, f->credit, q->quantum);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700390 q->inactive_flows--;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700391 }
Eric Dumazetf52ed892013-11-15 08:58:14 -0800392
393 /* Note: this overwrites f->age */
394 flow_queue_add(f, skb);
395
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700396 if (unlikely(f == &q->internal)) {
397 q->stat_internal_packets++;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700398 }
399 sch->q.qlen++;
400
401 return NET_XMIT_SUCCESS;
402}
403
404static void fq_check_throttled(struct fq_sched_data *q, u64 now)
405{
406 struct rb_node *p;
407
408 if (q->time_next_delayed_flow > now)
409 return;
410
411 q->time_next_delayed_flow = ~0ULL;
412 while ((p = rb_first(&q->delayed)) != NULL) {
413 struct fq_flow *f = container_of(p, struct fq_flow, rate_node);
414
415 if (f->time_next_packet > now) {
416 q->time_next_delayed_flow = f->time_next_packet;
417 break;
418 }
419 rb_erase(p, &q->delayed);
420 q->throttled_flows--;
421 fq_flow_add_tail(&q->old_flows, f);
422 }
423}
424
425static struct sk_buff *fq_dequeue(struct Qdisc *sch)
426{
427 struct fq_sched_data *q = qdisc_priv(sch);
Eric Dumazetd2de8752014-08-22 18:32:09 -0700428 u64 now = ktime_get_ns();
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700429 struct fq_flow_head *head;
430 struct sk_buff *skb;
431 struct fq_flow *f;
Eric Dumazet0eab5eb2013-10-01 09:10:16 -0700432 u32 rate;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700433
Eric Dumazet8d34ce12013-09-27 14:20:01 -0700434 skb = fq_dequeue_head(sch, &q->internal);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700435 if (skb)
436 goto out;
437 fq_check_throttled(q, now);
438begin:
439 head = &q->new_flows;
440 if (!head->first) {
441 head = &q->old_flows;
442 if (!head->first) {
443 if (q->time_next_delayed_flow != ~0ULL)
444 qdisc_watchdog_schedule_ns(&q->watchdog,
Eric Dumazetf2600cf2014-10-04 10:11:31 -0700445 q->time_next_delayed_flow,
446 false);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700447 return NULL;
448 }
449 }
450 f = head->first;
451
452 if (f->credit <= 0) {
453 f->credit += q->quantum;
454 head->first = f->next;
455 fq_flow_add_tail(&q->old_flows, f);
456 goto begin;
457 }
458
Eric Dumazet98781962015-02-03 18:31:53 -0800459 skb = f->head;
460 if (unlikely(skb && now < f->time_next_packet &&
461 !skb_is_tcp_pure_ack(skb))) {
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700462 head->first = f->next;
463 fq_flow_set_throttled(q, f);
464 goto begin;
465 }
466
Eric Dumazet8d34ce12013-09-27 14:20:01 -0700467 skb = fq_dequeue_head(sch, f);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700468 if (!skb) {
469 head->first = f->next;
470 /* force a pass through old_flows to prevent starvation */
471 if ((head == &q->new_flows) && q->old_flows.first) {
472 fq_flow_add_tail(&q->old_flows, f);
473 } else {
474 fq_flow_set_detached(f);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700475 q->inactive_flows++;
476 }
477 goto begin;
478 }
Eric Dumazet08f89b92013-08-30 09:46:43 -0700479 prefetch(&skb->end);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700480 f->credit -= qdisc_pkt_len(skb);
481
Eric Dumazet0eab5eb2013-10-01 09:10:16 -0700482 if (f->credit > 0 || !q->rate_enable)
483 goto out;
484
Eric Dumazet98781962015-02-03 18:31:53 -0800485 /* Do not pace locally generated ack packets */
486 if (skb_is_tcp_pure_ack(skb))
487 goto out;
488
Eric Dumazet7eec4172013-10-08 15:16:00 -0700489 rate = q->flow_max_rate;
Eric Dumazet86b3bfe2015-01-28 06:06:36 -0800490 if (skb->sk)
Eric Dumazet7eec4172013-10-08 15:16:00 -0700491 rate = min(skb->sk->sk_pacing_rate, rate);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700492
Eric Dumazet7eec4172013-10-08 15:16:00 -0700493 if (rate != ~0U) {
Eric Dumazet0eab5eb2013-10-01 09:10:16 -0700494 u32 plen = max(qdisc_pkt_len(skb), q->quantum);
495 u64 len = (u64)plen * NSEC_PER_SEC;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700496
Eric Dumazet7eec4172013-10-08 15:16:00 -0700497 if (likely(rate))
498 do_div(len, rate);
Eric Dumazet0eab5eb2013-10-01 09:10:16 -0700499 /* Since socket rate can change later,
Eric Dumazetced7a042014-11-25 08:57:29 -0800500 * clamp the delay to 1 second.
501 * Really, providers of too big packets should be fixed !
Eric Dumazet0eab5eb2013-10-01 09:10:16 -0700502 */
Eric Dumazetced7a042014-11-25 08:57:29 -0800503 if (unlikely(len > NSEC_PER_SEC)) {
504 len = NSEC_PER_SEC;
Eric Dumazet0eab5eb2013-10-01 09:10:16 -0700505 q->stat_pkts_too_long++;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700506 }
Eric Dumazet0eab5eb2013-10-01 09:10:16 -0700507
508 f->time_next_packet = now + len;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700509 }
510out:
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700511 qdisc_bstats_update(sch, skb);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700512 return skb;
513}
514
515static void fq_reset(struct Qdisc *sch)
516{
Eric Dumazet8d34ce12013-09-27 14:20:01 -0700517 struct fq_sched_data *q = qdisc_priv(sch);
518 struct rb_root *root;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700519 struct sk_buff *skb;
Eric Dumazet8d34ce12013-09-27 14:20:01 -0700520 struct rb_node *p;
521 struct fq_flow *f;
522 unsigned int idx;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700523
Eric Dumazet8d34ce12013-09-27 14:20:01 -0700524 while ((skb = fq_dequeue_head(sch, &q->internal)) != NULL)
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700525 kfree_skb(skb);
Eric Dumazet8d34ce12013-09-27 14:20:01 -0700526
527 if (!q->fq_root)
528 return;
529
530 for (idx = 0; idx < (1U << q->fq_trees_log); idx++) {
531 root = &q->fq_root[idx];
532 while ((p = rb_first(root)) != NULL) {
533 f = container_of(p, struct fq_flow, fq_node);
534 rb_erase(p, root);
535
536 while ((skb = fq_dequeue_head(sch, f)) != NULL)
537 kfree_skb(skb);
538
539 kmem_cache_free(fq_flow_cachep, f);
540 }
541 }
542 q->new_flows.first = NULL;
543 q->old_flows.first = NULL;
544 q->delayed = RB_ROOT;
545 q->flows = 0;
546 q->inactive_flows = 0;
547 q->throttled_flows = 0;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700548}
549
550static void fq_rehash(struct fq_sched_data *q,
551 struct rb_root *old_array, u32 old_log,
552 struct rb_root *new_array, u32 new_log)
553{
554 struct rb_node *op, **np, *parent;
555 struct rb_root *oroot, *nroot;
556 struct fq_flow *of, *nf;
557 int fcnt = 0;
558 u32 idx;
559
560 for (idx = 0; idx < (1U << old_log); idx++) {
561 oroot = &old_array[idx];
562 while ((op = rb_first(oroot)) != NULL) {
563 rb_erase(op, oroot);
564 of = container_of(op, struct fq_flow, fq_node);
565 if (fq_gc_candidate(of)) {
566 fcnt++;
567 kmem_cache_free(fq_flow_cachep, of);
568 continue;
569 }
570 nroot = &new_array[hash_32((u32)(long)of->sk, new_log)];
571
572 np = &nroot->rb_node;
573 parent = NULL;
574 while (*np) {
575 parent = *np;
576
577 nf = container_of(parent, struct fq_flow, fq_node);
578 BUG_ON(nf->sk == of->sk);
579
580 if (nf->sk > of->sk)
581 np = &parent->rb_right;
582 else
583 np = &parent->rb_left;
584 }
585
586 rb_link_node(&of->fq_node, parent, np);
587 rb_insert_color(&of->fq_node, nroot);
588 }
589 }
590 q->flows -= fcnt;
591 q->inactive_flows -= fcnt;
592 q->stat_gc_flows += fcnt;
593}
594
Eric Dumazetc3bd8542013-12-15 13:15:25 -0800595static void *fq_alloc_node(size_t sz, int node)
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700596{
Eric Dumazetc3bd8542013-12-15 13:15:25 -0800597 void *ptr;
598
599 ptr = kmalloc_node(sz, GFP_KERNEL | __GFP_REPEAT | __GFP_NOWARN, node);
600 if (!ptr)
601 ptr = vmalloc_node(sz, node);
602 return ptr;
603}
604
605static void fq_free(void *addr)
606{
WANG Cong4cb28972014-06-02 15:55:22 -0700607 kvfree(addr);
Eric Dumazetc3bd8542013-12-15 13:15:25 -0800608}
609
610static int fq_resize(struct Qdisc *sch, u32 log)
611{
612 struct fq_sched_data *q = qdisc_priv(sch);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700613 struct rb_root *array;
Eric Dumazet2d8d40a2014-03-06 22:57:52 -0800614 void *old_fq_root;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700615 u32 idx;
616
617 if (q->fq_root && log == q->fq_trees_log)
618 return 0;
619
Eric Dumazetc3bd8542013-12-15 13:15:25 -0800620 /* If XPS was setup, we can allocate memory on right NUMA node */
621 array = fq_alloc_node(sizeof(struct rb_root) << log,
622 netdev_queue_numa_node_read(sch->dev_queue));
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700623 if (!array)
624 return -ENOMEM;
625
626 for (idx = 0; idx < (1U << log); idx++)
627 array[idx] = RB_ROOT;
628
Eric Dumazet2d8d40a2014-03-06 22:57:52 -0800629 sch_tree_lock(sch);
630
631 old_fq_root = q->fq_root;
632 if (old_fq_root)
633 fq_rehash(q, old_fq_root, q->fq_trees_log, array, log);
634
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700635 q->fq_root = array;
636 q->fq_trees_log = log;
637
Eric Dumazet2d8d40a2014-03-06 22:57:52 -0800638 sch_tree_unlock(sch);
639
640 fq_free(old_fq_root);
641
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700642 return 0;
643}
644
645static const struct nla_policy fq_policy[TCA_FQ_MAX + 1] = {
646 [TCA_FQ_PLIMIT] = { .type = NLA_U32 },
647 [TCA_FQ_FLOW_PLIMIT] = { .type = NLA_U32 },
648 [TCA_FQ_QUANTUM] = { .type = NLA_U32 },
649 [TCA_FQ_INITIAL_QUANTUM] = { .type = NLA_U32 },
650 [TCA_FQ_RATE_ENABLE] = { .type = NLA_U32 },
651 [TCA_FQ_FLOW_DEFAULT_RATE] = { .type = NLA_U32 },
652 [TCA_FQ_FLOW_MAX_RATE] = { .type = NLA_U32 },
653 [TCA_FQ_BUCKETS_LOG] = { .type = NLA_U32 },
Eric Dumazetf52ed892013-11-15 08:58:14 -0800654 [TCA_FQ_FLOW_REFILL_DELAY] = { .type = NLA_U32 },
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700655};
656
657static int fq_change(struct Qdisc *sch, struct nlattr *opt)
658{
659 struct fq_sched_data *q = qdisc_priv(sch);
660 struct nlattr *tb[TCA_FQ_MAX + 1];
661 int err, drop_count = 0;
662 u32 fq_log;
663
664 if (!opt)
665 return -EINVAL;
666
667 err = nla_parse_nested(tb, TCA_FQ_MAX, opt, fq_policy);
668 if (err < 0)
669 return err;
670
671 sch_tree_lock(sch);
672
673 fq_log = q->fq_trees_log;
674
675 if (tb[TCA_FQ_BUCKETS_LOG]) {
676 u32 nval = nla_get_u32(tb[TCA_FQ_BUCKETS_LOG]);
677
678 if (nval >= 1 && nval <= ilog2(256*1024))
679 fq_log = nval;
680 else
681 err = -EINVAL;
682 }
683 if (tb[TCA_FQ_PLIMIT])
684 sch->limit = nla_get_u32(tb[TCA_FQ_PLIMIT]);
685
686 if (tb[TCA_FQ_FLOW_PLIMIT])
687 q->flow_plimit = nla_get_u32(tb[TCA_FQ_FLOW_PLIMIT]);
688
Kenneth Klette Jonassen3725a262015-02-03 17:49:18 +0100689 if (tb[TCA_FQ_QUANTUM]) {
690 u32 quantum = nla_get_u32(tb[TCA_FQ_QUANTUM]);
691
692 if (quantum > 0)
693 q->quantum = quantum;
694 else
695 err = -EINVAL;
696 }
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700697
698 if (tb[TCA_FQ_INITIAL_QUANTUM])
Eric Dumazetede869c2013-10-07 12:50:18 -0700699 q->initial_quantum = nla_get_u32(tb[TCA_FQ_INITIAL_QUANTUM]);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700700
701 if (tb[TCA_FQ_FLOW_DEFAULT_RATE])
Eric Dumazet65c5189a2013-11-15 08:57:26 -0800702 pr_warn_ratelimited("sch_fq: defrate %u ignored.\n",
703 nla_get_u32(tb[TCA_FQ_FLOW_DEFAULT_RATE]));
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700704
705 if (tb[TCA_FQ_FLOW_MAX_RATE])
706 q->flow_max_rate = nla_get_u32(tb[TCA_FQ_FLOW_MAX_RATE]);
707
708 if (tb[TCA_FQ_RATE_ENABLE]) {
709 u32 enable = nla_get_u32(tb[TCA_FQ_RATE_ENABLE]);
710
711 if (enable <= 1)
712 q->rate_enable = enable;
713 else
714 err = -EINVAL;
715 }
716
Eric Dumazetf52ed892013-11-15 08:58:14 -0800717 if (tb[TCA_FQ_FLOW_REFILL_DELAY]) {
718 u32 usecs_delay = nla_get_u32(tb[TCA_FQ_FLOW_REFILL_DELAY]) ;
719
720 q->flow_refill_delay = usecs_to_jiffies(usecs_delay);
721 }
722
Eric Dumazet06eb3952015-02-04 21:30:40 -0800723 if (tb[TCA_FQ_ORPHAN_MASK])
724 q->orphan_mask = nla_get_u32(tb[TCA_FQ_ORPHAN_MASK]);
725
Eric Dumazet2d8d40a2014-03-06 22:57:52 -0800726 if (!err) {
727 sch_tree_unlock(sch);
Eric Dumazetc3bd8542013-12-15 13:15:25 -0800728 err = fq_resize(sch, fq_log);
Eric Dumazet2d8d40a2014-03-06 22:57:52 -0800729 sch_tree_lock(sch);
730 }
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700731 while (sch->q.qlen > sch->limit) {
732 struct sk_buff *skb = fq_dequeue(sch);
733
Eric Dumazet8d34ce12013-09-27 14:20:01 -0700734 if (!skb)
735 break;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700736 kfree_skb(skb);
737 drop_count++;
738 }
739 qdisc_tree_decrease_qlen(sch, drop_count);
740
741 sch_tree_unlock(sch);
742 return err;
743}
744
745static void fq_destroy(struct Qdisc *sch)
746{
747 struct fq_sched_data *q = qdisc_priv(sch);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700748
Eric Dumazet8d34ce12013-09-27 14:20:01 -0700749 fq_reset(sch);
Eric Dumazetc3bd8542013-12-15 13:15:25 -0800750 fq_free(q->fq_root);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700751 qdisc_watchdog_cancel(&q->watchdog);
752}
753
754static int fq_init(struct Qdisc *sch, struct nlattr *opt)
755{
756 struct fq_sched_data *q = qdisc_priv(sch);
757 int err;
758
759 sch->limit = 10000;
760 q->flow_plimit = 100;
761 q->quantum = 2 * psched_mtu(qdisc_dev(sch));
762 q->initial_quantum = 10 * psched_mtu(qdisc_dev(sch));
Eric Dumazetf52ed892013-11-15 08:58:14 -0800763 q->flow_refill_delay = msecs_to_jiffies(40);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700764 q->flow_max_rate = ~0U;
765 q->rate_enable = 1;
766 q->new_flows.first = NULL;
767 q->old_flows.first = NULL;
768 q->delayed = RB_ROOT;
769 q->fq_root = NULL;
770 q->fq_trees_log = ilog2(1024);
Eric Dumazet06eb3952015-02-04 21:30:40 -0800771 q->orphan_mask = 1024 - 1;
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700772 qdisc_watchdog_init(&q->watchdog, sch);
773
774 if (opt)
775 err = fq_change(sch, opt);
776 else
Eric Dumazetc3bd8542013-12-15 13:15:25 -0800777 err = fq_resize(sch, q->fq_trees_log);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700778
779 return err;
780}
781
782static int fq_dump(struct Qdisc *sch, struct sk_buff *skb)
783{
784 struct fq_sched_data *q = qdisc_priv(sch);
785 struct nlattr *opts;
786
787 opts = nla_nest_start(skb, TCA_OPTIONS);
788 if (opts == NULL)
789 goto nla_put_failure;
790
Eric Dumazet65c5189a2013-11-15 08:57:26 -0800791 /* TCA_FQ_FLOW_DEFAULT_RATE is not used anymore */
792
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700793 if (nla_put_u32(skb, TCA_FQ_PLIMIT, sch->limit) ||
794 nla_put_u32(skb, TCA_FQ_FLOW_PLIMIT, q->flow_plimit) ||
795 nla_put_u32(skb, TCA_FQ_QUANTUM, q->quantum) ||
796 nla_put_u32(skb, TCA_FQ_INITIAL_QUANTUM, q->initial_quantum) ||
797 nla_put_u32(skb, TCA_FQ_RATE_ENABLE, q->rate_enable) ||
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700798 nla_put_u32(skb, TCA_FQ_FLOW_MAX_RATE, q->flow_max_rate) ||
Eric Dumazetf52ed892013-11-15 08:58:14 -0800799 nla_put_u32(skb, TCA_FQ_FLOW_REFILL_DELAY,
800 jiffies_to_usecs(q->flow_refill_delay)) ||
Eric Dumazet06eb3952015-02-04 21:30:40 -0800801 nla_put_u32(skb, TCA_FQ_ORPHAN_MASK, q->orphan_mask) ||
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700802 nla_put_u32(skb, TCA_FQ_BUCKETS_LOG, q->fq_trees_log))
803 goto nla_put_failure;
804
Yang Yingliangd59b7d82014-03-12 10:20:32 +0800805 return nla_nest_end(skb, opts);
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700806
807nla_put_failure:
808 return -1;
809}
810
811static int fq_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
812{
813 struct fq_sched_data *q = qdisc_priv(sch);
Eric Dumazetd2de8752014-08-22 18:32:09 -0700814 u64 now = ktime_get_ns();
Eric Dumazetafe4fd02013-08-29 15:49:55 -0700815 struct tc_fq_qd_stats st = {
816 .gc_flows = q->stat_gc_flows,
817 .highprio_packets = q->stat_internal_packets,
818 .tcp_retrans = q->stat_tcp_retrans,
819 .throttled = q->stat_throttled,
820 .flows_plimit = q->stat_flows_plimit,
821 .pkts_too_long = q->stat_pkts_too_long,
822 .allocation_errors = q->stat_allocation_errors,
823 .flows = q->flows,
824 .inactive_flows = q->inactive_flows,
825 .throttled_flows = q->throttled_flows,
826 .time_next_delayed_flow = q->time_next_delayed_flow - now,
827 };
828
829 return gnet_stats_copy_app(d, &st, sizeof(st));
830}
831
832static struct Qdisc_ops fq_qdisc_ops __read_mostly = {
833 .id = "fq",
834 .priv_size = sizeof(struct fq_sched_data),
835
836 .enqueue = fq_enqueue,
837 .dequeue = fq_dequeue,
838 .peek = qdisc_peek_dequeued,
839 .init = fq_init,
840 .reset = fq_reset,
841 .destroy = fq_destroy,
842 .change = fq_change,
843 .dump = fq_dump,
844 .dump_stats = fq_dump_stats,
845 .owner = THIS_MODULE,
846};
847
848static int __init fq_module_init(void)
849{
850 int ret;
851
852 fq_flow_cachep = kmem_cache_create("fq_flow_cache",
853 sizeof(struct fq_flow),
854 0, 0, NULL);
855 if (!fq_flow_cachep)
856 return -ENOMEM;
857
858 ret = register_qdisc(&fq_qdisc_ops);
859 if (ret)
860 kmem_cache_destroy(fq_flow_cachep);
861 return ret;
862}
863
864static void __exit fq_module_exit(void)
865{
866 unregister_qdisc(&fq_qdisc_ops);
867 kmem_cache_destroy(fq_flow_cachep);
868}
869
870module_init(fq_module_init)
871module_exit(fq_module_exit)
872MODULE_AUTHOR("Eric Dumazet");
873MODULE_LICENSE("GPL");