xref: /linux/net/sched/sch_cake.c (revision b2936b4fd56294e49d6c8e9152ea6c4982757c7d)
1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 
3 /* COMMON Applications Kept Enhanced (CAKE) discipline
4  *
5  * Copyright (C) 2014-2018 Jonathan Morton <chromatix99@gmail.com>
6  * Copyright (C) 2015-2018 Toke Høiland-Jørgensen <toke@toke.dk>
7  * Copyright (C) 2014-2018 Dave Täht <dave.taht@gmail.com>
8  * Copyright (C) 2015-2018 Sebastian Moeller <moeller0@gmx.de>
9  * (C) 2015-2018 Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
10  * Copyright (C) 2017-2018 Ryan Mounce <ryan@mounce.com.au>
11  *
12  * The CAKE Principles:
13  *		   (or, how to have your cake and eat it too)
14  *
15  * This is a combination of several shaping, AQM and FQ techniques into one
16  * easy-to-use package:
17  *
18  * - An overall bandwidth shaper, to move the bottleneck away from dumb CPE
19  *   equipment and bloated MACs.  This operates in deficit mode (as in sch_fq),
20  *   eliminating the need for any sort of burst parameter (eg. token bucket
21  *   depth).  Burst support is limited to that necessary to overcome scheduling
22  *   latency.
23  *
24  * - A Diffserv-aware priority queue, giving more priority to certain classes,
25  *   up to a specified fraction of bandwidth.  Above that bandwidth threshold,
26  *   the priority is reduced to avoid starving other tins.
27  *
28  * - Each priority tin has a separate Flow Queue system, to isolate traffic
29  *   flows from each other.  This prevents a burst on one flow from increasing
30  *   the delay to another.  Flows are distributed to queues using a
31  *   set-associative hash function.
32  *
33  * - Each queue is actively managed by Cobalt, which is a combination of the
34  *   Codel and Blue AQM algorithms.  This serves flows fairly, and signals
35  *   congestion early via ECN (if available) and/or packet drops, to keep
36  *   latency low.  The codel parameters are auto-tuned based on the bandwidth
37  *   setting, as is necessary at low bandwidths.
38  *
39  * The configuration parameters are kept deliberately simple for ease of use.
40  * Everything has sane defaults.  Complete generality of configuration is *not*
41  * a goal.
42  *
43  * The priority queue operates according to a weighted DRR scheme, combined with
44  * a bandwidth tracker which reuses the shaper logic to detect which side of the
45  * bandwidth sharing threshold the tin is operating.  This determines whether a
46  * priority-based weight (high) or a bandwidth-based weight (low) is used for
47  * that tin in the current pass.
48  *
49  * This qdisc was inspired by Eric Dumazet's fq_codel code, which he kindly
50  * granted us permission to leverage.
51  */
52 
53 #include <linux/module.h>
54 #include <linux/types.h>
55 #include <linux/kernel.h>
56 #include <linux/jiffies.h>
57 #include <linux/string.h>
58 #include <linux/in.h>
59 #include <linux/errno.h>
60 #include <linux/init.h>
61 #include <linux/skbuff.h>
62 #include <linux/jhash.h>
63 #include <linux/slab.h>
64 #include <linux/vmalloc.h>
65 #include <linux/reciprocal_div.h>
66 #include <net/netlink.h>
67 #include <linux/if_vlan.h>
68 #include <net/gso.h>
69 #include <net/pkt_sched.h>
70 #include <net/sch_priv.h>
71 #include <net/pkt_cls.h>
72 #include <net/tcp.h>
73 #include <net/flow_dissector.h>
74 
75 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
76 #include <net/netfilter/nf_conntrack_core.h>
77 #endif
78 
79 #define CAKE_SET_WAYS (8)
80 #define CAKE_MAX_TINS (8)
81 #define CAKE_QUEUES (1024)
82 #define CAKE_FLOW_MASK 63
83 #define CAKE_FLOW_NAT_FLAG 64
84 
85 /* struct cobalt_params - contains codel and blue parameters
86  * @interval:	codel initial drop rate
87  * @target:     maximum persistent sojourn time & blue update rate
88  * @mtu_time:   serialisation delay of maximum-size packet
89  * @p_inc:      increment of blue drop probability (0.32 fxp)
90  * @p_dec:      decrement of blue drop probability (0.32 fxp)
91  */
92 struct cobalt_params {
93 	u64	interval;
94 	u64	target;
95 	u64	mtu_time;
96 	u32	p_inc;
97 	u32	p_dec;
98 };
99 
100 /* struct cobalt_vars - contains codel and blue variables
101  * @count:		codel dropping frequency
102  * @rec_inv_sqrt:	reciprocal value of sqrt(count) >> 1
103  * @drop_next:		time to drop next packet, or when we dropped last
104  * @blue_timer:		Blue time to next drop
105  * @p_drop:		BLUE drop probability (0.32 fxp)
106  * @dropping:		set if in dropping state
107  * @ecn_marked:		set if marked
108  */
109 struct cobalt_vars {
110 	u32	count;
111 	u32	rec_inv_sqrt;
112 	ktime_t	drop_next;
113 	ktime_t	blue_timer;
114 	u32     p_drop;
115 	bool	dropping;
116 	bool    ecn_marked;
117 };
118 
119 enum {
120 	CAKE_SET_NONE = 0,
121 	CAKE_SET_SPARSE,
122 	CAKE_SET_SPARSE_WAIT, /* counted in SPARSE, actually in BULK */
123 	CAKE_SET_BULK,
124 	CAKE_SET_DECAYING
125 };
126 
127 struct cake_flow {
128 	/* this stuff is all needed per-flow at dequeue time */
129 	struct sk_buff	  *head;
130 	struct sk_buff	  *tail;
131 	struct list_head  flowchain;
132 	s32		  deficit;
133 	u32		  dropped;
134 	struct cobalt_vars cvars;
135 	u16		  srchost; /* index into cake_host table */
136 	u16		  dsthost;
137 	u8		  set;
138 }; /* please try to keep this structure <= 64 bytes */
139 
140 struct cake_host {
141 	u32 srchost_tag;
142 	u32 dsthost_tag;
143 	u16 srchost_bulk_flow_count;
144 	u16 dsthost_bulk_flow_count;
145 };
146 
147 struct cake_heap_entry {
148 	u16 t:3, b:10;
149 };
150 
151 struct cake_tin_data {
152 	struct cake_flow flows[CAKE_QUEUES];
153 	u32	backlogs[CAKE_QUEUES];
154 	u32	tags[CAKE_QUEUES]; /* for set association */
155 	u16	overflow_idx[CAKE_QUEUES];
156 	struct cake_host hosts[CAKE_QUEUES]; /* for triple isolation */
157 	u16	flow_quantum;
158 
159 	struct cobalt_params cparams;
160 	u32	drop_overlimit;
161 	u16	bulk_flow_count;
162 	u16	sparse_flow_count;
163 	u16	decaying_flow_count;
164 	u16	unresponsive_flow_count;
165 
166 	u32	max_skblen;
167 
168 	struct list_head new_flows;
169 	struct list_head old_flows;
170 	struct list_head decaying_flows;
171 
172 	/* time_next = time_this + ((len * rate_ns) >> rate_shft) */
173 	ktime_t	time_next_packet;
174 	u64	tin_rate_ns;
175 	u64	tin_rate_bps;
176 	u16	tin_rate_shft;
177 
178 	u16	tin_quantum;
179 	s32	tin_deficit;
180 	u32	tin_backlog;
181 	u32	tin_dropped;
182 	u32	tin_ecn_mark;
183 
184 	u32	packets;
185 	u64	bytes;
186 
187 	u32	ack_drops;
188 
189 	/* moving averages */
190 	u64 avge_delay;
191 	u64 peak_delay;
192 	u64 base_delay;
193 
194 	/* hash function stats */
195 	u32	way_directs;
196 	u32	way_hits;
197 	u32	way_misses;
198 	u32	way_collisions;
199 }; /* number of tins is small, so size of this struct doesn't matter much */
200 
201 struct cake_sched_config {
202 	u64		rate_bps;
203 	u64		interval;
204 	u64		target;
205 	u64		sync_time;
206 	u32		buffer_config_limit;
207 	u32		fwmark_mask;
208 	u16		fwmark_shft;
209 	s16		rate_overhead;
210 	u16		rate_mpu;
211 	u16		rate_flags;
212 	u8		tin_mode;
213 	u8		flow_mode;
214 	u8		atm_mode;
215 	u8		ack_filter;
216 	u8		is_shared;
217 };
218 
219 struct cake_sched_data {
220 	struct tcf_proto __rcu *filter_list; /* optional external classifier */
221 	struct tcf_block *block;
222 	struct cake_tin_data *tins;
223 	struct cake_sched_config *config;
224 	struct cake_sched_config initial_config;
225 
226 	struct cake_heap_entry overflow_heap[CAKE_QUEUES * CAKE_MAX_TINS];
227 
228 	/* time_next = time_this + ((len * rate_ns) >> rate_shft) */
229 	ktime_t		time_next_packet;
230 	ktime_t		failsafe_next_packet;
231 	u64		rate_ns;
232 	u16		rate_shft;
233 	u16		overflow_timeout;
234 	u16		tin_cnt;
235 
236 	/* resource tracking */
237 	u32		buffer_used;
238 	u32		buffer_max_used;
239 	u32		buffer_limit;
240 
241 	/* indices for dequeue */
242 	u16		cur_tin;
243 	u16		cur_flow;
244 
245 	struct qdisc_watchdog watchdog;
246 	const u8	*tin_index;
247 	const u8	*tin_order;
248 
249 	/* bandwidth capacity estimate */
250 	ktime_t		last_packet_time;
251 	ktime_t		avg_window_begin;
252 	u64		avg_packet_interval;
253 	u64		avg_window_bytes;
254 	u64		avg_peak_bandwidth;
255 	ktime_t		last_reconfig_time;
256 
257 	/* packet length stats */
258 	u32		avg_netoff;
259 	u16		max_netlen;
260 	u16		max_adjlen;
261 	u16		min_netlen;
262 	u16		min_adjlen;
263 
264 	/* mq sync state */
265 	u64		last_checked_active;
266 	u64		last_active;
267 	u32		active_queues;
268 };
269 
270 enum {
271 	CAKE_FLAG_OVERHEAD	   = BIT(0),
272 	CAKE_FLAG_AUTORATE_INGRESS = BIT(1),
273 	CAKE_FLAG_INGRESS	   = BIT(2),
274 	CAKE_FLAG_WASH		   = BIT(3),
275 	CAKE_FLAG_SPLIT_GSO	   = BIT(4)
276 };
277 
278 /* COBALT operates the Codel and BLUE algorithms in parallel, in order to
279  * obtain the best features of each.  Codel is excellent on flows which
280  * respond to congestion signals in a TCP-like way.  BLUE is more effective on
281  * unresponsive flows.
282  */
283 
284 struct cobalt_skb_cb {
285 	ktime_t enqueue_time;
286 	u32     adjusted_len;
287 };
288 
289 static u64 us_to_ns(u64 us)
290 {
291 	return us * NSEC_PER_USEC;
292 }
293 
294 static struct cobalt_skb_cb *get_cobalt_cb(const struct sk_buff *skb)
295 {
296 	qdisc_cb_private_validate(skb, sizeof(struct cobalt_skb_cb));
297 	return (struct cobalt_skb_cb *)qdisc_skb_cb(skb)->data;
298 }
299 
300 static ktime_t cobalt_get_enqueue_time(const struct sk_buff *skb)
301 {
302 	return get_cobalt_cb(skb)->enqueue_time;
303 }
304 
305 static void cobalt_set_enqueue_time(struct sk_buff *skb,
306 				    ktime_t now)
307 {
308 	get_cobalt_cb(skb)->enqueue_time = now;
309 }
310 
311 static u16 quantum_div[CAKE_QUEUES + 1] = {0};
312 
313 /* Diffserv lookup tables */
314 
315 static const u8 precedence[] = {
316 	0, 0, 0, 0, 0, 0, 0, 0,
317 	1, 1, 1, 1, 1, 1, 1, 1,
318 	2, 2, 2, 2, 2, 2, 2, 2,
319 	3, 3, 3, 3, 3, 3, 3, 3,
320 	4, 4, 4, 4, 4, 4, 4, 4,
321 	5, 5, 5, 5, 5, 5, 5, 5,
322 	6, 6, 6, 6, 6, 6, 6, 6,
323 	7, 7, 7, 7, 7, 7, 7, 7,
324 };
325 
326 static const u8 diffserv8[] = {
327 	2, 0, 1, 2, 4, 2, 2, 2,
328 	1, 2, 1, 2, 1, 2, 1, 2,
329 	5, 2, 4, 2, 4, 2, 4, 2,
330 	3, 2, 3, 2, 3, 2, 3, 2,
331 	6, 2, 3, 2, 3, 2, 3, 2,
332 	6, 2, 2, 2, 6, 2, 6, 2,
333 	7, 2, 2, 2, 2, 2, 2, 2,
334 	7, 2, 2, 2, 2, 2, 2, 2,
335 };
336 
337 static const u8 diffserv4[] = {
338 	0, 1, 0, 0, 2, 0, 0, 0,
339 	1, 0, 0, 0, 0, 0, 0, 0,
340 	2, 0, 2, 0, 2, 0, 2, 0,
341 	2, 0, 2, 0, 2, 0, 2, 0,
342 	3, 0, 2, 0, 2, 0, 2, 0,
343 	3, 0, 0, 0, 3, 0, 3, 0,
344 	3, 0, 0, 0, 0, 0, 0, 0,
345 	3, 0, 0, 0, 0, 0, 0, 0,
346 };
347 
348 static const u8 diffserv3[] = {
349 	0, 1, 0, 0, 2, 0, 0, 0,
350 	1, 0, 0, 0, 0, 0, 0, 0,
351 	0, 0, 0, 0, 0, 0, 0, 0,
352 	0, 0, 0, 0, 0, 0, 0, 0,
353 	0, 0, 0, 0, 0, 0, 0, 0,
354 	0, 0, 0, 0, 2, 0, 2, 0,
355 	2, 0, 0, 0, 0, 0, 0, 0,
356 	2, 0, 0, 0, 0, 0, 0, 0,
357 };
358 
359 static const u8 besteffort[] = {
360 	0, 0, 0, 0, 0, 0, 0, 0,
361 	0, 0, 0, 0, 0, 0, 0, 0,
362 	0, 0, 0, 0, 0, 0, 0, 0,
363 	0, 0, 0, 0, 0, 0, 0, 0,
364 	0, 0, 0, 0, 0, 0, 0, 0,
365 	0, 0, 0, 0, 0, 0, 0, 0,
366 	0, 0, 0, 0, 0, 0, 0, 0,
367 	0, 0, 0, 0, 0, 0, 0, 0,
368 };
369 
370 /* tin priority order for stats dumping */
371 
372 static const u8 normal_order[] = {0, 1, 2, 3, 4, 5, 6, 7};
373 static const u8 bulk_order[] = {1, 0, 2, 3};
374 
375 /* There is a big difference in timing between the accurate values placed in the
376  * cache and the approximations given by a single Newton step for small count
377  * values, particularly when stepping from count 1 to 2 or vice versa. Hence,
378  * these values are calculated using eight Newton steps, using the
379  * implementation below. Above 16, a single Newton step gives sufficient
380  * accuracy in either direction, given the precision stored.
381  *
382  * The magnitude of the error when stepping up to count 2 is such as to give the
383  * value that *should* have been produced at count 4.
384  */
385 
386 #define REC_INV_SQRT_CACHE (16)
387 static const u32 inv_sqrt_cache[REC_INV_SQRT_CACHE] = {
388 		~0,         ~0, 3037000500, 2479700525,
389 	2147483647, 1920767767, 1753413056, 1623345051,
390 	1518500250, 1431655765, 1358187914, 1294981364,
391 	1239850263, 1191209601, 1147878294, 1108955788
392 };
393 
394 static void cake_set_rate(struct cake_tin_data *b, u64 rate, u32 mtu,
395 			  u64 target_ns, u64 rtt_est_ns);
396 /* http://en.wikipedia.org/wiki/Methods_of_computing_square_roots
397  * new_invsqrt = (invsqrt / 2) * (3 - count * invsqrt^2)
398  *
399  * Here, invsqrt is a fixed point number (< 1.0), 32bit mantissa, aka Q0.32
400  */
401 
402 static void cobalt_newton_step(struct cobalt_vars *vars)
403 {
404 	u32 invsqrt, invsqrt2;
405 	u64 val;
406 
407 	invsqrt = vars->rec_inv_sqrt;
408 	invsqrt2 = ((u64)invsqrt * invsqrt) >> 32;
409 	val = (3LL << 32) - ((u64)vars->count * invsqrt2);
410 
411 	val >>= 2; /* avoid overflow in following multiply */
412 	val = (val * invsqrt) >> (32 - 2 + 1);
413 
414 	vars->rec_inv_sqrt = val;
415 }
416 
417 static void cobalt_invsqrt(struct cobalt_vars *vars)
418 {
419 	if (vars->count < REC_INV_SQRT_CACHE)
420 		vars->rec_inv_sqrt = inv_sqrt_cache[vars->count];
421 	else
422 		cobalt_newton_step(vars);
423 }
424 
425 static void cobalt_vars_init(struct cobalt_vars *vars)
426 {
427 	memset(vars, 0, sizeof(*vars));
428 }
429 
430 /* CoDel control_law is t + interval/sqrt(count)
431  * We maintain in rec_inv_sqrt the reciprocal value of sqrt(count) to avoid
432  * both sqrt() and divide operation.
433  */
434 static ktime_t cobalt_control(ktime_t t,
435 			      u64 interval,
436 			      u32 rec_inv_sqrt)
437 {
438 	return ktime_add_ns(t, reciprocal_scale(interval,
439 						rec_inv_sqrt));
440 }
441 
442 /* Call this when a packet had to be dropped due to queue overflow.  Returns
443  * true if the BLUE state was quiescent before but active after this call.
444  */
445 static bool cobalt_queue_full(struct cobalt_vars *vars,
446 			      struct cobalt_params *p,
447 			      ktime_t now)
448 {
449 	bool up = false;
450 
451 	if (ktime_to_ns(ktime_sub(now, vars->blue_timer)) > p->target) {
452 		up = !vars->p_drop;
453 		vars->p_drop += p->p_inc;
454 		if (vars->p_drop < p->p_inc)
455 			vars->p_drop = ~0;
456 		vars->blue_timer = now;
457 	}
458 	vars->dropping = true;
459 	vars->drop_next = now;
460 	if (!vars->count)
461 		vars->count = 1;
462 
463 	return up;
464 }
465 
466 /* Call this when the queue was serviced but turned out to be empty.  Returns
467  * true if the BLUE state was active before but quiescent after this call.
468  */
469 static bool cobalt_queue_empty(struct cobalt_vars *vars,
470 			       struct cobalt_params *p,
471 			       ktime_t now)
472 {
473 	bool down = false;
474 
475 	if (vars->p_drop &&
476 	    ktime_to_ns(ktime_sub(now, vars->blue_timer)) > p->target) {
477 		if (vars->p_drop < p->p_dec)
478 			vars->p_drop = 0;
479 		else
480 			vars->p_drop -= p->p_dec;
481 		vars->blue_timer = now;
482 		down = !vars->p_drop;
483 	}
484 	vars->dropping = false;
485 
486 	if (vars->count && ktime_to_ns(ktime_sub(now, vars->drop_next)) >= 0) {
487 		vars->count--;
488 		cobalt_invsqrt(vars);
489 		vars->drop_next = cobalt_control(vars->drop_next,
490 						 p->interval,
491 						 vars->rec_inv_sqrt);
492 	}
493 
494 	return down;
495 }
496 
497 /* Call this with a freshly dequeued packet for possible congestion marking.
498  * Returns true as an instruction to drop the packet, false for delivery.
499  */
500 static enum skb_drop_reason cobalt_should_drop(struct cobalt_vars *vars,
501 					       struct cobalt_params *p,
502 					       ktime_t now,
503 					       struct sk_buff *skb,
504 					       u32 bulk_flows)
505 {
506 	enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
507 	bool next_due, over_target;
508 	ktime_t schedule;
509 	u64 sojourn;
510 
511 /* The 'schedule' variable records, in its sign, whether 'now' is before or
512  * after 'drop_next'.  This allows 'drop_next' to be updated before the next
513  * scheduling decision is actually branched, without destroying that
514  * information.  Similarly, the first 'schedule' value calculated is preserved
515  * in the boolean 'next_due'.
516  *
517  * As for 'drop_next', we take advantage of the fact that 'interval' is both
518  * the delay between first exceeding 'target' and the first signalling event,
519  * *and* the scaling factor for the signalling frequency.  It's therefore very
520  * natural to use a single mechanism for both purposes, and eliminates a
521  * significant amount of reference Codel's spaghetti code.  To help with this,
522  * both the '0' and '1' entries in the invsqrt cache are 0xFFFFFFFF, as close
523  * as possible to 1.0 in fixed-point.
524  */
525 
526 	sojourn = ktime_to_ns(ktime_sub(now, cobalt_get_enqueue_time(skb)));
527 	schedule = ktime_sub(now, vars->drop_next);
528 	over_target = sojourn > p->target &&
529 		      sojourn > p->mtu_time * bulk_flows * 2 &&
530 		      sojourn > p->mtu_time * 4;
531 	next_due = vars->count && ktime_to_ns(schedule) >= 0;
532 
533 	vars->ecn_marked = false;
534 
535 	if (over_target) {
536 		if (!vars->dropping) {
537 			vars->dropping = true;
538 			vars->drop_next = cobalt_control(now,
539 							 p->interval,
540 							 vars->rec_inv_sqrt);
541 		}
542 		if (!vars->count)
543 			vars->count = 1;
544 	} else if (vars->dropping) {
545 		vars->dropping = false;
546 	}
547 
548 	if (next_due && vars->dropping) {
549 		/* Use ECN mark if possible, otherwise drop */
550 		if (!(vars->ecn_marked = INET_ECN_set_ce(skb)))
551 			reason = SKB_DROP_REASON_QDISC_CONGESTED;
552 
553 		vars->count++;
554 		if (!vars->count)
555 			vars->count--;
556 		cobalt_invsqrt(vars);
557 		vars->drop_next = cobalt_control(vars->drop_next,
558 						 p->interval,
559 						 vars->rec_inv_sqrt);
560 		schedule = ktime_sub(now, vars->drop_next);
561 	} else {
562 		while (next_due) {
563 			vars->count--;
564 			cobalt_invsqrt(vars);
565 			vars->drop_next = cobalt_control(vars->drop_next,
566 							 p->interval,
567 							 vars->rec_inv_sqrt);
568 			schedule = ktime_sub(now, vars->drop_next);
569 			next_due = vars->count && ktime_to_ns(schedule) >= 0;
570 		}
571 	}
572 
573 	/* Simple BLUE implementation.  Lack of ECN is deliberate. */
574 	if (vars->p_drop && reason == SKB_NOT_DROPPED_YET &&
575 	    get_random_u32() < vars->p_drop)
576 		reason = SKB_DROP_REASON_CAKE_FLOOD;
577 
578 	/* Overload the drop_next field as an activity timeout */
579 	if (!vars->count)
580 		vars->drop_next = ktime_add_ns(now, p->interval);
581 	else if (ktime_to_ns(schedule) > 0 && reason == SKB_NOT_DROPPED_YET)
582 		vars->drop_next = now;
583 
584 	return reason;
585 }
586 
587 static bool cake_update_flowkeys(struct flow_keys *keys,
588 				 const struct sk_buff *skb)
589 {
590 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
591 	struct nf_conntrack_tuple tuple = {};
592 	bool rev = !skb->_nfct, upd = false;
593 	__be32 ip;
594 
595 	if (skb_protocol(skb, true) != htons(ETH_P_IP))
596 		return false;
597 
598 	if (!nf_ct_get_tuple_skb(&tuple, skb))
599 		return false;
600 
601 	ip = rev ? tuple.dst.u3.ip : tuple.src.u3.ip;
602 	if (ip != keys->addrs.v4addrs.src) {
603 		keys->addrs.v4addrs.src = ip;
604 		upd = true;
605 	}
606 	ip = rev ? tuple.src.u3.ip : tuple.dst.u3.ip;
607 	if (ip != keys->addrs.v4addrs.dst) {
608 		keys->addrs.v4addrs.dst = ip;
609 		upd = true;
610 	}
611 
612 	if (keys->ports.ports) {
613 		__be16 port;
614 
615 		port = rev ? tuple.dst.u.all : tuple.src.u.all;
616 		if (port != keys->ports.src) {
617 			keys->ports.src = port;
618 			upd = true;
619 		}
620 		port = rev ? tuple.src.u.all : tuple.dst.u.all;
621 		if (port != keys->ports.dst) {
622 			port = keys->ports.dst;
623 			upd = true;
624 		}
625 	}
626 	return upd;
627 #else
628 	return false;
629 #endif
630 }
631 
632 /* Cake has several subtle multiple bit settings. In these cases you
633  *  would be matching triple isolate mode as well.
634  */
635 
636 static bool cake_dsrc(int flow_mode)
637 {
638 	return (flow_mode & CAKE_FLOW_DUAL_SRC) == CAKE_FLOW_DUAL_SRC;
639 }
640 
641 static bool cake_ddst(int flow_mode)
642 {
643 	return (flow_mode & CAKE_FLOW_DUAL_DST) == CAKE_FLOW_DUAL_DST;
644 }
645 
646 static void cake_dec_srchost_bulk_flow_count(struct cake_tin_data *q,
647 					     struct cake_flow *flow,
648 					     int flow_mode)
649 {
650 	if (likely(cake_dsrc(flow_mode) &&
651 		   q->hosts[flow->srchost].srchost_bulk_flow_count))
652 		q->hosts[flow->srchost].srchost_bulk_flow_count--;
653 }
654 
655 static void cake_inc_srchost_bulk_flow_count(struct cake_tin_data *q,
656 					     struct cake_flow *flow,
657 					     int flow_mode)
658 {
659 	if (likely(cake_dsrc(flow_mode) &&
660 		   q->hosts[flow->srchost].srchost_bulk_flow_count < CAKE_QUEUES))
661 		q->hosts[flow->srchost].srchost_bulk_flow_count++;
662 }
663 
664 static void cake_dec_dsthost_bulk_flow_count(struct cake_tin_data *q,
665 					     struct cake_flow *flow,
666 					     int flow_mode)
667 {
668 	if (likely(cake_ddst(flow_mode) &&
669 		   q->hosts[flow->dsthost].dsthost_bulk_flow_count))
670 		q->hosts[flow->dsthost].dsthost_bulk_flow_count--;
671 }
672 
673 static void cake_inc_dsthost_bulk_flow_count(struct cake_tin_data *q,
674 					     struct cake_flow *flow,
675 					     int flow_mode)
676 {
677 	if (likely(cake_ddst(flow_mode) &&
678 		   q->hosts[flow->dsthost].dsthost_bulk_flow_count < CAKE_QUEUES))
679 		q->hosts[flow->dsthost].dsthost_bulk_flow_count++;
680 }
681 
682 static u16 cake_get_flow_quantum(struct cake_tin_data *q,
683 				 struct cake_flow *flow,
684 				 int flow_mode)
685 {
686 	u16 host_load = 1;
687 
688 	if (cake_dsrc(flow_mode))
689 		host_load = max(host_load,
690 				q->hosts[flow->srchost].srchost_bulk_flow_count);
691 
692 	if (cake_ddst(flow_mode))
693 		host_load = max(host_load,
694 				q->hosts[flow->dsthost].dsthost_bulk_flow_count);
695 
696 	/* The get_random_u16() is a way to apply dithering to avoid
697 	 * accumulating roundoff errors
698 	 */
699 	return (q->flow_quantum * quantum_div[host_load] +
700 		get_random_u16()) >> 16;
701 }
702 
703 static u32 cake_hash(struct cake_tin_data *q, const struct sk_buff *skb,
704 		     int flow_mode, u16 flow_override, u16 host_override)
705 {
706 	bool hash_flows = (!flow_override && !!(flow_mode & CAKE_FLOW_FLOWS));
707 	bool hash_hosts = (!host_override && !!(flow_mode & CAKE_FLOW_HOSTS));
708 	bool nat_enabled = !!(flow_mode & CAKE_FLOW_NAT_FLAG);
709 	u32 flow_hash = 0, srchost_hash = 0, dsthost_hash = 0;
710 	u16 reduced_hash, srchost_idx, dsthost_idx;
711 	struct flow_keys keys, host_keys;
712 	bool use_skbhash = skb->l4_hash;
713 
714 	if (unlikely(flow_mode == CAKE_FLOW_NONE))
715 		return 0;
716 
717 	/* If both overrides are set, or we can use the SKB hash and nat mode is
718 	 * disabled, we can skip packet dissection entirely. If nat mode is
719 	 * enabled there's another check below after doing the conntrack lookup.
720 	 */
721 	if ((!hash_flows || (use_skbhash && !nat_enabled)) && !hash_hosts)
722 		goto skip_hash;
723 
724 	skb_flow_dissect_flow_keys(skb, &keys,
725 				   FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL);
726 
727 	/* Don't use the SKB hash if we change the lookup keys from conntrack */
728 	if (nat_enabled && cake_update_flowkeys(&keys, skb))
729 		use_skbhash = false;
730 
731 	/* If we can still use the SKB hash and don't need the host hash, we can
732 	 * skip the rest of the hashing procedure
733 	 */
734 	if (use_skbhash && !hash_hosts)
735 		goto skip_hash;
736 
737 	/* flow_hash_from_keys() sorts the addresses by value, so we have
738 	 * to preserve their order in a separate data structure to treat
739 	 * src and dst host addresses as independently selectable.
740 	 */
741 	host_keys = keys;
742 	host_keys.ports.ports     = 0;
743 	host_keys.basic.ip_proto  = 0;
744 	host_keys.keyid.keyid     = 0;
745 	host_keys.tags.flow_label = 0;
746 
747 	switch (host_keys.control.addr_type) {
748 	case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
749 		host_keys.addrs.v4addrs.src = 0;
750 		dsthost_hash = flow_hash_from_keys(&host_keys);
751 		host_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
752 		host_keys.addrs.v4addrs.dst = 0;
753 		srchost_hash = flow_hash_from_keys(&host_keys);
754 		break;
755 
756 	case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
757 		memset(&host_keys.addrs.v6addrs.src, 0,
758 		       sizeof(host_keys.addrs.v6addrs.src));
759 		dsthost_hash = flow_hash_from_keys(&host_keys);
760 		host_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
761 		memset(&host_keys.addrs.v6addrs.dst, 0,
762 		       sizeof(host_keys.addrs.v6addrs.dst));
763 		srchost_hash = flow_hash_from_keys(&host_keys);
764 		break;
765 
766 	default:
767 		dsthost_hash = 0;
768 		srchost_hash = 0;
769 	}
770 
771 	/* This *must* be after the above switch, since as a
772 	 * side-effect it sorts the src and dst addresses.
773 	 */
774 	if (hash_flows && !use_skbhash)
775 		flow_hash = flow_hash_from_keys(&keys);
776 
777 skip_hash:
778 	if (flow_override)
779 		flow_hash = flow_override - 1;
780 	else if (use_skbhash && (flow_mode & CAKE_FLOW_FLOWS))
781 		flow_hash = skb->hash;
782 	if (host_override) {
783 		dsthost_hash = host_override - 1;
784 		srchost_hash = host_override - 1;
785 	}
786 
787 	if (!(flow_mode & CAKE_FLOW_FLOWS)) {
788 		if (flow_mode & CAKE_FLOW_SRC_IP)
789 			flow_hash ^= srchost_hash;
790 
791 		if (flow_mode & CAKE_FLOW_DST_IP)
792 			flow_hash ^= dsthost_hash;
793 	}
794 
795 	reduced_hash = flow_hash % CAKE_QUEUES;
796 
797 	/* set-associative hashing */
798 	/* fast path if no hash collision (direct lookup succeeds) */
799 	if (likely(q->tags[reduced_hash] == flow_hash &&
800 		   q->flows[reduced_hash].set)) {
801 		q->way_directs++;
802 	} else {
803 		u32 inner_hash = reduced_hash % CAKE_SET_WAYS;
804 		u32 outer_hash = reduced_hash - inner_hash;
805 		bool allocate_src = false;
806 		bool allocate_dst = false;
807 		u32 i, k;
808 
809 		/* check if any active queue in the set is reserved for
810 		 * this flow.
811 		 */
812 		for (i = 0, k = inner_hash; i < CAKE_SET_WAYS;
813 		     i++, k = (k + 1) % CAKE_SET_WAYS) {
814 			if (q->tags[outer_hash + k] == flow_hash) {
815 				if (i)
816 					q->way_hits++;
817 
818 				if (!q->flows[outer_hash + k].set) {
819 					/* need to increment host refcnts */
820 					allocate_src = cake_dsrc(flow_mode);
821 					allocate_dst = cake_ddst(flow_mode);
822 				}
823 
824 				goto found;
825 			}
826 		}
827 
828 		/* no queue is reserved for this flow, look for an
829 		 * empty one.
830 		 */
831 		for (i = 0; i < CAKE_SET_WAYS;
832 			 i++, k = (k + 1) % CAKE_SET_WAYS) {
833 			if (!q->flows[outer_hash + k].set) {
834 				q->way_misses++;
835 				allocate_src = cake_dsrc(flow_mode);
836 				allocate_dst = cake_ddst(flow_mode);
837 				goto found;
838 			}
839 		}
840 
841 		/* With no empty queues, default to the original
842 		 * queue, accept the collision, update the host tags.
843 		 */
844 		q->way_collisions++;
845 		allocate_src = cake_dsrc(flow_mode);
846 		allocate_dst = cake_ddst(flow_mode);
847 
848 		if (q->flows[outer_hash + k].set == CAKE_SET_BULK) {
849 			cake_dec_srchost_bulk_flow_count(q, &q->flows[outer_hash + k], flow_mode);
850 			cake_dec_dsthost_bulk_flow_count(q, &q->flows[outer_hash + k], flow_mode);
851 		}
852 found:
853 		/* reserve queue for future packets in same flow */
854 		reduced_hash = outer_hash + k;
855 		q->tags[reduced_hash] = flow_hash;
856 
857 		if (allocate_src) {
858 			srchost_idx = srchost_hash % CAKE_QUEUES;
859 			inner_hash = srchost_idx % CAKE_SET_WAYS;
860 			outer_hash = srchost_idx - inner_hash;
861 			for (i = 0, k = inner_hash; i < CAKE_SET_WAYS;
862 				i++, k = (k + 1) % CAKE_SET_WAYS) {
863 				if (q->hosts[outer_hash + k].srchost_tag ==
864 				    srchost_hash)
865 					goto found_src;
866 			}
867 			for (i = 0; i < CAKE_SET_WAYS;
868 				i++, k = (k + 1) % CAKE_SET_WAYS) {
869 				if (!q->hosts[outer_hash + k].srchost_bulk_flow_count)
870 					break;
871 			}
872 			q->hosts[outer_hash + k].srchost_tag = srchost_hash;
873 found_src:
874 			srchost_idx = outer_hash + k;
875 			q->flows[reduced_hash].srchost = srchost_idx;
876 
877 			if (q->flows[reduced_hash].set == CAKE_SET_BULK)
878 				cake_inc_srchost_bulk_flow_count(q, &q->flows[reduced_hash], flow_mode);
879 		}
880 
881 		if (allocate_dst) {
882 			dsthost_idx = dsthost_hash % CAKE_QUEUES;
883 			inner_hash = dsthost_idx % CAKE_SET_WAYS;
884 			outer_hash = dsthost_idx - inner_hash;
885 			for (i = 0, k = inner_hash; i < CAKE_SET_WAYS;
886 			     i++, k = (k + 1) % CAKE_SET_WAYS) {
887 				if (q->hosts[outer_hash + k].dsthost_tag ==
888 				    dsthost_hash)
889 					goto found_dst;
890 			}
891 			for (i = 0; i < CAKE_SET_WAYS;
892 			     i++, k = (k + 1) % CAKE_SET_WAYS) {
893 				if (!q->hosts[outer_hash + k].dsthost_bulk_flow_count)
894 					break;
895 			}
896 			q->hosts[outer_hash + k].dsthost_tag = dsthost_hash;
897 found_dst:
898 			dsthost_idx = outer_hash + k;
899 			q->flows[reduced_hash].dsthost = dsthost_idx;
900 
901 			if (q->flows[reduced_hash].set == CAKE_SET_BULK)
902 				cake_inc_dsthost_bulk_flow_count(q, &q->flows[reduced_hash], flow_mode);
903 		}
904 	}
905 
906 	return reduced_hash;
907 }
908 
909 /* helper functions : might be changed when/if skb use a standard list_head */
910 /* remove one skb from head of slot queue */
911 
912 static struct sk_buff *dequeue_head(struct cake_flow *flow)
913 {
914 	struct sk_buff *skb = flow->head;
915 
916 	if (skb) {
917 		flow->head = skb->next;
918 		skb_mark_not_on_list(skb);
919 	}
920 
921 	return skb;
922 }
923 
924 /* add skb to flow queue (tail add) */
925 
926 static void flow_queue_add(struct cake_flow *flow, struct sk_buff *skb)
927 {
928 	if (!flow->head)
929 		flow->head = skb;
930 	else
931 		flow->tail->next = skb;
932 	flow->tail = skb;
933 	skb->next = NULL;
934 }
935 
936 static struct iphdr *cake_get_iphdr(const struct sk_buff *skb,
937 				    struct ipv6hdr *buf)
938 {
939 	unsigned int offset = skb_network_offset(skb);
940 	struct iphdr *iph;
941 
942 	iph = skb_header_pointer(skb, offset, sizeof(struct iphdr), buf);
943 
944 	if (!iph)
945 		return NULL;
946 
947 	if (iph->version == 4 && iph->protocol == IPPROTO_IPV6)
948 		return skb_header_pointer(skb, offset + iph->ihl * 4,
949 					  sizeof(struct ipv6hdr), buf);
950 
951 	else if (iph->version == 4)
952 		return iph;
953 
954 	else if (iph->version == 6)
955 		return skb_header_pointer(skb, offset, sizeof(struct ipv6hdr),
956 					  buf);
957 
958 	return NULL;
959 }
960 
961 static struct tcphdr *cake_get_tcphdr(const struct sk_buff *skb,
962 				      void *buf, unsigned int bufsize)
963 {
964 	unsigned int offset = skb_network_offset(skb);
965 	const struct ipv6hdr *ipv6h;
966 	const struct tcphdr *tcph;
967 	const struct iphdr *iph;
968 	struct ipv6hdr _ipv6h;
969 	struct tcphdr _tcph;
970 
971 	ipv6h = skb_header_pointer(skb, offset, sizeof(_ipv6h), &_ipv6h);
972 
973 	if (!ipv6h)
974 		return NULL;
975 
976 	if (ipv6h->version == 4) {
977 		iph = (struct iphdr *)ipv6h;
978 		offset += iph->ihl * 4;
979 
980 		/* special-case 6in4 tunnelling, as that is a common way to get
981 		 * v6 connectivity in the home
982 		 */
983 		if (iph->protocol == IPPROTO_IPV6) {
984 			ipv6h = skb_header_pointer(skb, offset,
985 						   sizeof(_ipv6h), &_ipv6h);
986 
987 			if (!ipv6h || ipv6h->nexthdr != IPPROTO_TCP)
988 				return NULL;
989 
990 			offset += sizeof(struct ipv6hdr);
991 
992 		} else if (iph->protocol != IPPROTO_TCP) {
993 			return NULL;
994 		}
995 
996 	} else if (ipv6h->version == 6) {
997 		if (ipv6h->nexthdr != IPPROTO_TCP)
998 			return NULL;
999 
1000 		offset += sizeof(struct ipv6hdr);
1001 	} else {
1002 		return NULL;
1003 	}
1004 
1005 	tcph = skb_header_pointer(skb, offset, sizeof(_tcph), &_tcph);
1006 	if (!tcph || tcph->doff < 5)
1007 		return NULL;
1008 
1009 	return skb_header_pointer(skb, offset,
1010 				  min(__tcp_hdrlen(tcph), bufsize), buf);
1011 }
1012 
1013 static const void *cake_get_tcpopt(const struct tcphdr *tcph,
1014 				   int code, int *oplen)
1015 {
1016 	/* inspired by tcp_parse_options in tcp_input.c */
1017 	int length = __tcp_hdrlen(tcph) - sizeof(struct tcphdr);
1018 	const u8 *ptr = (const u8 *)(tcph + 1);
1019 
1020 	while (length > 0) {
1021 		int opcode = *ptr++;
1022 		int opsize;
1023 
1024 		if (opcode == TCPOPT_EOL)
1025 			break;
1026 		if (opcode == TCPOPT_NOP) {
1027 			length--;
1028 			continue;
1029 		}
1030 		if (length < 2)
1031 			break;
1032 		opsize = *ptr++;
1033 		if (opsize < 2 || opsize > length)
1034 			break;
1035 
1036 		if (opcode == code) {
1037 			*oplen = opsize;
1038 			return ptr;
1039 		}
1040 
1041 		ptr += opsize - 2;
1042 		length -= opsize;
1043 	}
1044 
1045 	return NULL;
1046 }
1047 
1048 /* Compare two SACK sequences. A sequence is considered greater if it SACKs more
1049  * bytes than the other. In the case where both sequences ACKs bytes that the
1050  * other doesn't, A is considered greater. DSACKs in A also makes A be
1051  * considered greater.
1052  *
1053  * @return -1, 0 or 1 as normal compare functions
1054  */
1055 static int cake_tcph_sack_compare(const struct tcphdr *tcph_a,
1056 				  const struct tcphdr *tcph_b)
1057 {
1058 	const struct tcp_sack_block_wire *sack_a, *sack_b;
1059 	u32 ack_seq_a = ntohl(tcph_a->ack_seq);
1060 	u32 bytes_a = 0, bytes_b = 0;
1061 	int oplen_a, oplen_b;
1062 	bool first = true;
1063 
1064 	sack_a = cake_get_tcpopt(tcph_a, TCPOPT_SACK, &oplen_a);
1065 	sack_b = cake_get_tcpopt(tcph_b, TCPOPT_SACK, &oplen_b);
1066 
1067 	/* pointers point to option contents */
1068 	oplen_a -= TCPOLEN_SACK_BASE;
1069 	oplen_b -= TCPOLEN_SACK_BASE;
1070 
1071 	if (sack_a && oplen_a >= sizeof(*sack_a) &&
1072 	    (!sack_b || oplen_b < sizeof(*sack_b)))
1073 		return -1;
1074 	else if (sack_b && oplen_b >= sizeof(*sack_b) &&
1075 		 (!sack_a || oplen_a < sizeof(*sack_a)))
1076 		return 1;
1077 	else if ((!sack_a || oplen_a < sizeof(*sack_a)) &&
1078 		 (!sack_b || oplen_b < sizeof(*sack_b)))
1079 		return 0;
1080 
1081 	while (oplen_a >= sizeof(*sack_a)) {
1082 		const struct tcp_sack_block_wire *sack_tmp = sack_b;
1083 		u32 start_a = get_unaligned_be32(&sack_a->start_seq);
1084 		u32 end_a = get_unaligned_be32(&sack_a->end_seq);
1085 		int oplen_tmp = oplen_b;
1086 		bool found = false;
1087 
1088 		/* DSACK; always considered greater to prevent dropping */
1089 		if (before(start_a, ack_seq_a))
1090 			return -1;
1091 
1092 		bytes_a += end_a - start_a;
1093 
1094 		while (oplen_tmp >= sizeof(*sack_tmp)) {
1095 			u32 start_b = get_unaligned_be32(&sack_tmp->start_seq);
1096 			u32 end_b = get_unaligned_be32(&sack_tmp->end_seq);
1097 
1098 			/* first time through we count the total size */
1099 			if (first)
1100 				bytes_b += end_b - start_b;
1101 
1102 			if (!after(start_b, start_a) && !before(end_b, end_a)) {
1103 				found = true;
1104 				if (!first)
1105 					break;
1106 			}
1107 			oplen_tmp -= sizeof(*sack_tmp);
1108 			sack_tmp++;
1109 		}
1110 
1111 		if (!found)
1112 			return -1;
1113 
1114 		oplen_a -= sizeof(*sack_a);
1115 		sack_a++;
1116 		first = false;
1117 	}
1118 
1119 	/* If we made it this far, all ranges SACKed by A are covered by B, so
1120 	 * either the SACKs are equal, or B SACKs more bytes.
1121 	 */
1122 	return bytes_b > bytes_a ? 1 : 0;
1123 }
1124 
1125 static void cake_tcph_get_tstamp(const struct tcphdr *tcph,
1126 				 u32 *tsval, u32 *tsecr)
1127 {
1128 	const u8 *ptr;
1129 	int opsize;
1130 
1131 	ptr = cake_get_tcpopt(tcph, TCPOPT_TIMESTAMP, &opsize);
1132 
1133 	if (ptr && opsize == TCPOLEN_TIMESTAMP) {
1134 		*tsval = get_unaligned_be32(ptr);
1135 		*tsecr = get_unaligned_be32(ptr + 4);
1136 	}
1137 }
1138 
1139 static bool cake_tcph_may_drop(const struct tcphdr *tcph,
1140 			       u32 tstamp_new, u32 tsecr_new)
1141 {
1142 	/* inspired by tcp_parse_options in tcp_input.c */
1143 	int length = __tcp_hdrlen(tcph) - sizeof(struct tcphdr);
1144 	const u8 *ptr = (const u8 *)(tcph + 1);
1145 	u32 tstamp, tsecr;
1146 
1147 	/* 3 reserved flags must be unset to avoid future breakage
1148 	 * ACK must be set
1149 	 * ECE/CWR are handled separately
1150 	 * All other flags URG/PSH/RST/SYN/FIN must be unset
1151 	 * 0x0FFF0000 = all TCP flags (confirm ACK=1, others zero)
1152 	 * 0x00C00000 = CWR/ECE (handled separately)
1153 	 * 0x0F3F0000 = 0x0FFF0000 & ~0x00C00000
1154 	 */
1155 	if (((tcp_flag_word(tcph) &
1156 	      cpu_to_be32(0x0F3F0000)) != TCP_FLAG_ACK))
1157 		return false;
1158 
1159 	while (length > 0) {
1160 		int opcode = *ptr++;
1161 		int opsize;
1162 
1163 		if (opcode == TCPOPT_EOL)
1164 			break;
1165 		if (opcode == TCPOPT_NOP) {
1166 			length--;
1167 			continue;
1168 		}
1169 		if (length < 2)
1170 			break;
1171 		opsize = *ptr++;
1172 		if (opsize < 2 || opsize > length)
1173 			break;
1174 
1175 		switch (opcode) {
1176 		case TCPOPT_MD5SIG: /* doesn't influence state */
1177 			break;
1178 
1179 		case TCPOPT_SACK: /* stricter checking performed later */
1180 			if (opsize % 8 != 2)
1181 				return false;
1182 			break;
1183 
1184 		case TCPOPT_TIMESTAMP:
1185 			/* only drop timestamps lower than new */
1186 			if (opsize != TCPOLEN_TIMESTAMP)
1187 				return false;
1188 			tstamp = get_unaligned_be32(ptr);
1189 			tsecr = get_unaligned_be32(ptr + 4);
1190 			if (after(tstamp, tstamp_new) ||
1191 			    after(tsecr, tsecr_new))
1192 				return false;
1193 			break;
1194 
1195 		case TCPOPT_MSS:  /* these should only be set on SYN */
1196 		case TCPOPT_WINDOW:
1197 		case TCPOPT_SACK_PERM:
1198 		case TCPOPT_FASTOPEN:
1199 		case TCPOPT_EXP:
1200 		default: /* don't drop if any unknown options are present */
1201 			return false;
1202 		}
1203 
1204 		ptr += opsize - 2;
1205 		length -= opsize;
1206 	}
1207 
1208 	return true;
1209 }
1210 
1211 static struct sk_buff *cake_ack_filter(struct cake_sched_data *q,
1212 				       struct cake_flow *flow)
1213 {
1214 	bool aggressive = q->config->ack_filter == CAKE_ACK_AGGRESSIVE;
1215 	struct sk_buff *elig_ack = NULL, *elig_ack_prev = NULL;
1216 	struct sk_buff *skb_check, *skb_prev = NULL;
1217 	const struct ipv6hdr *ipv6h, *ipv6h_check;
1218 	unsigned char _tcph[64], _tcph_check[64];
1219 	const struct tcphdr *tcph, *tcph_check;
1220 	const struct iphdr *iph, *iph_check;
1221 	struct ipv6hdr _iph, _iph_check;
1222 	const struct sk_buff *skb;
1223 	int seglen, num_found = 0;
1224 	u32 tstamp = 0, tsecr = 0;
1225 	__be32 elig_flags = 0;
1226 	int sack_comp;
1227 
1228 	/* no other possible ACKs to filter */
1229 	if (flow->head == flow->tail)
1230 		return NULL;
1231 
1232 	skb = flow->tail;
1233 	tcph = cake_get_tcphdr(skb, _tcph, sizeof(_tcph));
1234 	iph = cake_get_iphdr(skb, &_iph);
1235 	if (!tcph)
1236 		return NULL;
1237 
1238 	cake_tcph_get_tstamp(tcph, &tstamp, &tsecr);
1239 
1240 	/* the 'triggering' packet need only have the ACK flag set.
1241 	 * also check that SYN is not set, as there won't be any previous ACKs.
1242 	 */
1243 	if ((tcp_flag_word(tcph) &
1244 	     (TCP_FLAG_ACK | TCP_FLAG_SYN)) != TCP_FLAG_ACK)
1245 		return NULL;
1246 
1247 	/* the 'triggering' ACK is at the tail of the queue, we have already
1248 	 * returned if it is the only packet in the flow. loop through the rest
1249 	 * of the queue looking for pure ACKs with the same 5-tuple as the
1250 	 * triggering one.
1251 	 */
1252 	for (skb_check = flow->head;
1253 	     skb_check && skb_check != skb;
1254 	     skb_prev = skb_check, skb_check = skb_check->next) {
1255 		iph_check = cake_get_iphdr(skb_check, &_iph_check);
1256 		tcph_check = cake_get_tcphdr(skb_check, &_tcph_check,
1257 					     sizeof(_tcph_check));
1258 
1259 		/* only TCP packets with matching 5-tuple are eligible, and only
1260 		 * drop safe headers
1261 		 */
1262 		if (!tcph_check || iph->version != iph_check->version ||
1263 		    tcph_check->source != tcph->source ||
1264 		    tcph_check->dest != tcph->dest)
1265 			continue;
1266 
1267 		if (iph_check->version == 4) {
1268 			if (iph_check->saddr != iph->saddr ||
1269 			    iph_check->daddr != iph->daddr)
1270 				continue;
1271 
1272 			seglen = iph_totlen(skb, iph_check) -
1273 				       (4 * iph_check->ihl);
1274 		} else if (iph_check->version == 6) {
1275 			ipv6h = (struct ipv6hdr *)iph;
1276 			ipv6h_check = (struct ipv6hdr *)iph_check;
1277 
1278 			if (ipv6_addr_cmp(&ipv6h_check->saddr, &ipv6h->saddr) ||
1279 			    ipv6_addr_cmp(&ipv6h_check->daddr, &ipv6h->daddr))
1280 				continue;
1281 
1282 			seglen = ipv6_payload_len(skb, ipv6h_check);
1283 		} else {
1284 			WARN_ON(1);  /* shouldn't happen */
1285 			continue;
1286 		}
1287 
1288 		/* If the ECE/CWR flags changed from the previous eligible
1289 		 * packet in the same flow, we should no longer be dropping that
1290 		 * previous packet as this would lose information.
1291 		 */
1292 		if (elig_ack && (tcp_flag_word(tcph_check) &
1293 				 (TCP_FLAG_ECE | TCP_FLAG_CWR)) != elig_flags) {
1294 			elig_ack = NULL;
1295 			elig_ack_prev = NULL;
1296 			num_found--;
1297 		}
1298 
1299 		/* Check TCP options and flags, don't drop ACKs with segment
1300 		 * data, and don't drop ACKs with a higher cumulative ACK
1301 		 * counter than the triggering packet. Check ACK seqno here to
1302 		 * avoid parsing SACK options of packets we are going to exclude
1303 		 * anyway.
1304 		 */
1305 		if (!cake_tcph_may_drop(tcph_check, tstamp, tsecr) ||
1306 		    (seglen - __tcp_hdrlen(tcph_check)) != 0 ||
1307 		    after(ntohl(tcph_check->ack_seq), ntohl(tcph->ack_seq)))
1308 			continue;
1309 
1310 		/* Check SACK options. The triggering packet must SACK more data
1311 		 * than the ACK under consideration, or SACK the same range but
1312 		 * have a larger cumulative ACK counter. The latter is a
1313 		 * pathological case, but is contained in the following check
1314 		 * anyway, just to be safe.
1315 		 */
1316 		sack_comp = cake_tcph_sack_compare(tcph_check, tcph);
1317 
1318 		if (sack_comp < 0 ||
1319 		    (ntohl(tcph_check->ack_seq) == ntohl(tcph->ack_seq) &&
1320 		     sack_comp == 0))
1321 			continue;
1322 
1323 		/* At this point we have found an eligible pure ACK to drop; if
1324 		 * we are in aggressive mode, we are done. Otherwise, keep
1325 		 * searching unless this is the second eligible ACK we
1326 		 * found.
1327 		 *
1328 		 * Since we want to drop ACK closest to the head of the queue,
1329 		 * save the first eligible ACK we find, even if we need to loop
1330 		 * again.
1331 		 */
1332 		if (!elig_ack) {
1333 			elig_ack = skb_check;
1334 			elig_ack_prev = skb_prev;
1335 			elig_flags = (tcp_flag_word(tcph_check)
1336 				      & (TCP_FLAG_ECE | TCP_FLAG_CWR));
1337 		}
1338 
1339 		if (num_found++ > 0)
1340 			goto found;
1341 	}
1342 
1343 	/* We made it through the queue without finding two eligible ACKs . If
1344 	 * we found a single eligible ACK we can drop it in aggressive mode if
1345 	 * we can guarantee that this does not interfere with ECN flag
1346 	 * information. We ensure this by dropping it only if the enqueued
1347 	 * packet is consecutive with the eligible ACK, and their flags match.
1348 	 */
1349 	if (elig_ack && aggressive && elig_ack->next == skb &&
1350 	    (elig_flags == (tcp_flag_word(tcph) &
1351 			    (TCP_FLAG_ECE | TCP_FLAG_CWR))))
1352 		goto found;
1353 
1354 	return NULL;
1355 
1356 found:
1357 	if (elig_ack_prev)
1358 		elig_ack_prev->next = elig_ack->next;
1359 	else
1360 		flow->head = elig_ack->next;
1361 
1362 	skb_mark_not_on_list(elig_ack);
1363 
1364 	return elig_ack;
1365 }
1366 
1367 static u64 cake_ewma(u64 avg, u64 sample, u32 shift)
1368 {
1369 	avg -= avg >> shift;
1370 	avg += sample >> shift;
1371 	return avg;
1372 }
1373 
1374 static u32 cake_calc_overhead(struct cake_sched_data *qd, u32 len, u32 off)
1375 {
1376 	struct cake_sched_config *q = qd->config;
1377 
1378 	if (q->rate_flags & CAKE_FLAG_OVERHEAD)
1379 		len -= off;
1380 
1381 	if (qd->max_netlen < len)
1382 		qd->max_netlen = len;
1383 	if (qd->min_netlen > len)
1384 		qd->min_netlen = len;
1385 
1386 	len += q->rate_overhead;
1387 
1388 	if (len < q->rate_mpu)
1389 		len = q->rate_mpu;
1390 
1391 	if (q->atm_mode == CAKE_ATM_ATM) {
1392 		len += 47;
1393 		len /= 48;
1394 		len *= 53;
1395 	} else if (q->atm_mode == CAKE_ATM_PTM) {
1396 		/* Add one byte per 64 bytes or part thereof.
1397 		 * This is conservative and easier to calculate than the
1398 		 * precise value.
1399 		 */
1400 		len += (len + 63) / 64;
1401 	}
1402 
1403 	if (qd->max_adjlen < len)
1404 		qd->max_adjlen = len;
1405 	if (qd->min_adjlen > len)
1406 		qd->min_adjlen = len;
1407 
1408 	return len;
1409 }
1410 
1411 static u32 cake_overhead(struct cake_sched_data *q, const struct sk_buff *skb)
1412 {
1413 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
1414 	unsigned int hdr_len, last_len = 0;
1415 	u32 off = skb_network_offset(skb);
1416 	u16 segs = qdisc_pkt_segs(skb);
1417 	u32 len = qdisc_pkt_len(skb);
1418 
1419 	q->avg_netoff = cake_ewma(q->avg_netoff, off << 16, 8);
1420 
1421 	if (segs == 1)
1422 		return cake_calc_overhead(q, len, off);
1423 
1424 	/* borrowed from qdisc_pkt_len_segs_init() */
1425 	if (!skb->encapsulation)
1426 		hdr_len = skb_transport_offset(skb);
1427 	else
1428 		hdr_len = skb_inner_transport_offset(skb);
1429 
1430 	/* + transport layer */
1431 	if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 |
1432 						SKB_GSO_TCPV6))) {
1433 		const struct tcphdr *th;
1434 		struct tcphdr _tcphdr;
1435 
1436 		th = skb_header_pointer(skb, hdr_len,
1437 					sizeof(_tcphdr), &_tcphdr);
1438 		if (likely(th))
1439 			hdr_len += __tcp_hdrlen(th);
1440 	} else {
1441 		struct udphdr _udphdr;
1442 
1443 		if (skb_header_pointer(skb, hdr_len,
1444 				       sizeof(_udphdr), &_udphdr))
1445 			hdr_len += sizeof(struct udphdr);
1446 	}
1447 
1448 	len = shinfo->gso_size + hdr_len;
1449 	last_len = skb->len - shinfo->gso_size * (segs - 1);
1450 
1451 	return (cake_calc_overhead(q, len, off) * (segs - 1) +
1452 		cake_calc_overhead(q, last_len, off));
1453 }
1454 
1455 static void cake_heap_swap(struct cake_sched_data *q, u16 i, u16 j)
1456 {
1457 	struct cake_heap_entry ii = q->overflow_heap[i];
1458 	struct cake_heap_entry jj = q->overflow_heap[j];
1459 
1460 	q->overflow_heap[i] = jj;
1461 	q->overflow_heap[j] = ii;
1462 
1463 	q->tins[ii.t].overflow_idx[ii.b] = j;
1464 	q->tins[jj.t].overflow_idx[jj.b] = i;
1465 }
1466 
1467 static u32 cake_heap_get_backlog(const struct cake_sched_data *q, u16 i)
1468 {
1469 	struct cake_heap_entry ii = q->overflow_heap[i];
1470 
1471 	return q->tins[ii.t].backlogs[ii.b];
1472 }
1473 
1474 static void cake_heapify(struct cake_sched_data *q, u16 i)
1475 {
1476 	static const u32 a = CAKE_MAX_TINS * CAKE_QUEUES;
1477 	u32 mb = cake_heap_get_backlog(q, i);
1478 	u32 m = i;
1479 
1480 	while (m < a) {
1481 		u32 l = m + m + 1;
1482 		u32 r = l + 1;
1483 
1484 		if (l < a) {
1485 			u32 lb = cake_heap_get_backlog(q, l);
1486 
1487 			if (lb > mb) {
1488 				m  = l;
1489 				mb = lb;
1490 			}
1491 		}
1492 
1493 		if (r < a) {
1494 			u32 rb = cake_heap_get_backlog(q, r);
1495 
1496 			if (rb > mb) {
1497 				m  = r;
1498 				mb = rb;
1499 			}
1500 		}
1501 
1502 		if (m != i) {
1503 			cake_heap_swap(q, i, m);
1504 			i = m;
1505 		} else {
1506 			break;
1507 		}
1508 	}
1509 }
1510 
1511 static void cake_heapify_up(struct cake_sched_data *q, u16 i)
1512 {
1513 	while (i > 0 && i < CAKE_MAX_TINS * CAKE_QUEUES) {
1514 		u16 p = (i - 1) >> 1;
1515 		u32 ib = cake_heap_get_backlog(q, i);
1516 		u32 pb = cake_heap_get_backlog(q, p);
1517 
1518 		if (ib > pb) {
1519 			cake_heap_swap(q, i, p);
1520 			i = p;
1521 		} else {
1522 			break;
1523 		}
1524 	}
1525 }
1526 
1527 static int cake_advance_shaper(struct cake_sched_data *q,
1528 			       struct cake_tin_data *b,
1529 			       struct sk_buff *skb,
1530 			       ktime_t now, bool drop)
1531 {
1532 	u32 len = get_cobalt_cb(skb)->adjusted_len;
1533 
1534 	/* charge packet bandwidth to this tin
1535 	 * and to the global shaper.
1536 	 */
1537 	if (q->rate_ns) {
1538 		u64 tin_dur = (len * b->tin_rate_ns) >> b->tin_rate_shft;
1539 		u64 global_dur = (len * q->rate_ns) >> q->rate_shft;
1540 		u64 failsafe_dur = global_dur + (global_dur >> 1);
1541 
1542 		if (ktime_before(b->time_next_packet, now))
1543 			b->time_next_packet = ktime_add_ns(b->time_next_packet,
1544 							   tin_dur);
1545 
1546 		else if (ktime_before(b->time_next_packet,
1547 				      ktime_add_ns(now, tin_dur)))
1548 			b->time_next_packet = ktime_add_ns(now, tin_dur);
1549 
1550 		q->time_next_packet = ktime_add_ns(q->time_next_packet,
1551 						   global_dur);
1552 		if (!drop)
1553 			q->failsafe_next_packet = \
1554 				ktime_add_ns(q->failsafe_next_packet,
1555 					     failsafe_dur);
1556 	}
1557 	return len;
1558 }
1559 
1560 static unsigned int cake_drop(struct Qdisc *sch, struct sk_buff **to_free)
1561 {
1562 	struct cake_sched_data *q = qdisc_priv(sch);
1563 	ktime_t now = ktime_get();
1564 	u32 idx = 0, tin = 0, len;
1565 	struct cake_heap_entry qq;
1566 	struct cake_tin_data *b;
1567 	struct cake_flow *flow;
1568 	struct sk_buff *skb;
1569 
1570 	if (!q->overflow_timeout) {
1571 		int i;
1572 		/* Build fresh max-heap */
1573 		for (i = CAKE_MAX_TINS * CAKE_QUEUES / 2 - 1; i >= 0; i--)
1574 			cake_heapify(q, i);
1575 	}
1576 	q->overflow_timeout = 65535;
1577 
1578 	/* select longest queue for pruning */
1579 	qq  = q->overflow_heap[0];
1580 	tin = qq.t;
1581 	idx = qq.b;
1582 
1583 	b = &q->tins[tin];
1584 	flow = &b->flows[idx];
1585 	skb = dequeue_head(flow);
1586 	if (unlikely(!skb)) {
1587 		/* heap has gone wrong, rebuild it next time */
1588 		q->overflow_timeout = 0;
1589 		return idx + (tin << 16);
1590 	}
1591 
1592 	if (cobalt_queue_full(&flow->cvars, &b->cparams, now))
1593 		b->unresponsive_flow_count++;
1594 
1595 	len = qdisc_pkt_len(skb);
1596 	q->buffer_used      -= skb->truesize;
1597 	b->backlogs[idx]    -= len;
1598 	b->tin_backlog      -= len;
1599 	sch->qstats.backlog -= len;
1600 
1601 	flow->dropped++;
1602 	b->tin_dropped++;
1603 
1604 	if (q->config->rate_flags & CAKE_FLAG_INGRESS)
1605 		cake_advance_shaper(q, b, skb, now, true);
1606 
1607 	qdisc_drop_reason(skb, sch, to_free, SKB_DROP_REASON_QDISC_OVERLIMIT);
1608 	sch->q.qlen--;
1609 
1610 	cake_heapify(q, 0);
1611 
1612 	return idx + (tin << 16);
1613 }
1614 
1615 static u8 cake_handle_diffserv(struct sk_buff *skb, bool wash)
1616 {
1617 	const int offset = skb_network_offset(skb);
1618 	u16 *buf, buf_;
1619 	u8 dscp;
1620 
1621 	switch (skb_protocol(skb, true)) {
1622 	case htons(ETH_P_IP):
1623 		buf = skb_header_pointer(skb, offset, sizeof(buf_), &buf_);
1624 		if (unlikely(!buf))
1625 			return 0;
1626 
1627 		/* ToS is in the second byte of iphdr */
1628 		dscp = ipv4_get_dsfield((struct iphdr *)buf) >> 2;
1629 
1630 		if (wash && dscp) {
1631 			const int wlen = offset + sizeof(struct iphdr);
1632 
1633 			if (!pskb_may_pull(skb, wlen) ||
1634 			    skb_try_make_writable(skb, wlen))
1635 				return 0;
1636 
1637 			ipv4_change_dsfield(ip_hdr(skb), INET_ECN_MASK, 0);
1638 		}
1639 
1640 		return dscp;
1641 
1642 	case htons(ETH_P_IPV6):
1643 		buf = skb_header_pointer(skb, offset, sizeof(buf_), &buf_);
1644 		if (unlikely(!buf))
1645 			return 0;
1646 
1647 		/* Traffic class is in the first and second bytes of ipv6hdr */
1648 		dscp = ipv6_get_dsfield((struct ipv6hdr *)buf) >> 2;
1649 
1650 		if (wash && dscp) {
1651 			const int wlen = offset + sizeof(struct ipv6hdr);
1652 
1653 			if (!pskb_may_pull(skb, wlen) ||
1654 			    skb_try_make_writable(skb, wlen))
1655 				return 0;
1656 
1657 			ipv6_change_dsfield(ipv6_hdr(skb), INET_ECN_MASK, 0);
1658 		}
1659 
1660 		return dscp;
1661 
1662 	case htons(ETH_P_ARP):
1663 		return 0x38;  /* CS7 - Net Control */
1664 
1665 	default:
1666 		/* If there is no Diffserv field, treat as best-effort */
1667 		return 0;
1668 	}
1669 }
1670 
1671 static struct cake_tin_data *cake_select_tin(struct Qdisc *sch,
1672 					     struct sk_buff *skb)
1673 {
1674 	struct cake_sched_data *qd = qdisc_priv(sch);
1675 	struct cake_sched_config *q = qd->config;
1676 	u32 tin, mark;
1677 	bool wash;
1678 	u8 dscp;
1679 
1680 	/* Tin selection: Default to diffserv-based selection, allow overriding
1681 	 * using firewall marks or skb->priority. Call DSCP parsing early if
1682 	 * wash is enabled, otherwise defer to below to skip unneeded parsing.
1683 	 */
1684 	mark = (skb->mark & q->fwmark_mask) >> q->fwmark_shft;
1685 	wash = !!(q->rate_flags & CAKE_FLAG_WASH);
1686 	if (wash)
1687 		dscp = cake_handle_diffserv(skb, wash);
1688 
1689 	if (q->tin_mode == CAKE_DIFFSERV_BESTEFFORT)
1690 		tin = 0;
1691 
1692 	else if (mark && mark <= qd->tin_cnt)
1693 		tin = qd->tin_order[mark - 1];
1694 
1695 	else if (TC_H_MAJ(skb->priority) == sch->handle &&
1696 		 TC_H_MIN(skb->priority) > 0 &&
1697 		 TC_H_MIN(skb->priority) <= qd->tin_cnt)
1698 		tin = qd->tin_order[TC_H_MIN(skb->priority) - 1];
1699 
1700 	else {
1701 		if (!wash)
1702 			dscp = cake_handle_diffserv(skb, wash);
1703 		tin = qd->tin_index[dscp];
1704 
1705 		if (unlikely(tin >= qd->tin_cnt))
1706 			tin = 0;
1707 	}
1708 
1709 	return &qd->tins[tin];
1710 }
1711 
1712 static u32 cake_classify(struct Qdisc *sch, struct cake_tin_data **t,
1713 			 struct sk_buff *skb, int flow_mode, int *qerr)
1714 {
1715 	struct cake_sched_data *q = qdisc_priv(sch);
1716 	struct tcf_proto *filter;
1717 	struct tcf_result res;
1718 	u16 flow = 0, host = 0;
1719 	int result;
1720 
1721 	filter = rcu_dereference_bh(q->filter_list);
1722 	if (!filter)
1723 		goto hash;
1724 
1725 	*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
1726 	result = tcf_classify(skb, NULL, filter, &res, false);
1727 
1728 	if (result >= 0) {
1729 #ifdef CONFIG_NET_CLS_ACT
1730 		switch (result) {
1731 		case TC_ACT_STOLEN:
1732 		case TC_ACT_QUEUED:
1733 		case TC_ACT_TRAP:
1734 			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
1735 			fallthrough;
1736 		case TC_ACT_SHOT:
1737 			return 0;
1738 		}
1739 #endif
1740 		if (TC_H_MIN(res.classid) <= CAKE_QUEUES)
1741 			flow = TC_H_MIN(res.classid);
1742 		if (TC_H_MAJ(res.classid) <= (CAKE_QUEUES << 16))
1743 			host = TC_H_MAJ(res.classid) >> 16;
1744 	}
1745 hash:
1746 	*t = cake_select_tin(sch, skb);
1747 	return cake_hash(*t, skb, flow_mode, flow, host) + 1;
1748 }
1749 
1750 static void cake_reconfigure(struct Qdisc *sch);
1751 
1752 static s32 cake_enqueue(struct sk_buff *skb, struct Qdisc *sch,
1753 			struct sk_buff **to_free)
1754 {
1755 	u32 idx, tin, prev_qlen, prev_backlog, drop_id;
1756 	struct cake_sched_data *q = qdisc_priv(sch);
1757 	int len = qdisc_pkt_len(skb), ret;
1758 	struct sk_buff *ack = NULL;
1759 	ktime_t now = ktime_get();
1760 	struct cake_tin_data *b;
1761 	struct cake_flow *flow;
1762 	bool same_flow = false;
1763 
1764 	/* choose flow to insert into */
1765 	idx = cake_classify(sch, &b, skb, q->config->flow_mode, &ret);
1766 	if (idx == 0) {
1767 		if (ret & __NET_XMIT_BYPASS)
1768 			qdisc_qstats_drop(sch);
1769 		__qdisc_drop(skb, to_free);
1770 		return ret;
1771 	}
1772 	tin = (u32)(b - q->tins);
1773 	idx--;
1774 	flow = &b->flows[idx];
1775 
1776 	/* ensure shaper state isn't stale */
1777 	if (!b->tin_backlog) {
1778 		if (ktime_before(b->time_next_packet, now))
1779 			b->time_next_packet = now;
1780 
1781 		if (!sch->q.qlen) {
1782 			if (ktime_before(q->time_next_packet, now)) {
1783 				q->failsafe_next_packet = now;
1784 				q->time_next_packet = now;
1785 			} else if (ktime_after(q->time_next_packet, now) &&
1786 				   ktime_after(q->failsafe_next_packet, now)) {
1787 				u64 next = \
1788 					min(ktime_to_ns(q->time_next_packet),
1789 					    ktime_to_ns(
1790 						   q->failsafe_next_packet));
1791 				sch->qstats.overlimits++;
1792 				qdisc_watchdog_schedule_ns(&q->watchdog, next);
1793 			}
1794 		}
1795 	}
1796 
1797 	if (unlikely(len > b->max_skblen))
1798 		b->max_skblen = len;
1799 
1800 	if (qdisc_pkt_segs(skb) > 1 && q->config->rate_flags & CAKE_FLAG_SPLIT_GSO) {
1801 		struct sk_buff *segs, *nskb;
1802 		netdev_features_t features = netif_skb_features(skb);
1803 		unsigned int slen = 0, numsegs = 0;
1804 
1805 		segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
1806 		if (IS_ERR_OR_NULL(segs))
1807 			return qdisc_drop(skb, sch, to_free);
1808 
1809 		skb_list_walk_safe(segs, segs, nskb) {
1810 			skb_mark_not_on_list(segs);
1811 			qdisc_skb_cb(segs)->pkt_len = segs->len;
1812 			qdisc_skb_cb(segs)->pkt_segs = 1;
1813 			cobalt_set_enqueue_time(segs, now);
1814 			get_cobalt_cb(segs)->adjusted_len = cake_overhead(q,
1815 									  segs);
1816 			flow_queue_add(flow, segs);
1817 
1818 			sch->q.qlen++;
1819 			numsegs++;
1820 			slen += segs->len;
1821 			q->buffer_used += segs->truesize;
1822 			b->packets++;
1823 		}
1824 
1825 		/* stats */
1826 		b->bytes	    += slen;
1827 		b->backlogs[idx]    += slen;
1828 		b->tin_backlog      += slen;
1829 		sch->qstats.backlog += slen;
1830 		q->avg_window_bytes += slen;
1831 
1832 		qdisc_tree_reduce_backlog(sch, 1-numsegs, len-slen);
1833 		consume_skb(skb);
1834 	} else {
1835 		/* not splitting */
1836 		int ack_pkt_len = 0;
1837 
1838 		cobalt_set_enqueue_time(skb, now);
1839 		get_cobalt_cb(skb)->adjusted_len = cake_overhead(q, skb);
1840 		flow_queue_add(flow, skb);
1841 
1842 		if (q->config->ack_filter)
1843 			ack = cake_ack_filter(q, flow);
1844 
1845 		if (ack) {
1846 			b->ack_drops++;
1847 			sch->qstats.drops++;
1848 			ack_pkt_len = qdisc_pkt_len(ack);
1849 			b->bytes += ack_pkt_len;
1850 			q->buffer_used += skb->truesize - ack->truesize;
1851 			if (q->config->rate_flags & CAKE_FLAG_INGRESS)
1852 				cake_advance_shaper(q, b, ack, now, true);
1853 
1854 			qdisc_tree_reduce_backlog(sch, 1, ack_pkt_len);
1855 			consume_skb(ack);
1856 		} else {
1857 			sch->q.qlen++;
1858 			q->buffer_used      += skb->truesize;
1859 		}
1860 
1861 		/* stats */
1862 		b->packets++;
1863 		b->bytes	    += len - ack_pkt_len;
1864 		b->backlogs[idx]    += len - ack_pkt_len;
1865 		b->tin_backlog      += len - ack_pkt_len;
1866 		sch->qstats.backlog += len - ack_pkt_len;
1867 		q->avg_window_bytes += len - ack_pkt_len;
1868 	}
1869 
1870 	if (q->overflow_timeout)
1871 		cake_heapify_up(q, b->overflow_idx[idx]);
1872 
1873 	/* incoming bandwidth capacity estimate */
1874 	if (q->config->rate_flags & CAKE_FLAG_AUTORATE_INGRESS) {
1875 		u64 packet_interval = \
1876 			ktime_to_ns(ktime_sub(now, q->last_packet_time));
1877 
1878 		if (packet_interval > NSEC_PER_SEC)
1879 			packet_interval = NSEC_PER_SEC;
1880 
1881 		/* filter out short-term bursts, eg. wifi aggregation */
1882 		q->avg_packet_interval = \
1883 			cake_ewma(q->avg_packet_interval,
1884 				  packet_interval,
1885 				  (packet_interval > q->avg_packet_interval ?
1886 					  2 : 8));
1887 
1888 		q->last_packet_time = now;
1889 
1890 		if (packet_interval > q->avg_packet_interval) {
1891 			u64 window_interval = \
1892 				ktime_to_ns(ktime_sub(now,
1893 						      q->avg_window_begin));
1894 			u64 b = q->avg_window_bytes * (u64)NSEC_PER_SEC;
1895 
1896 			b = div64_u64(b, window_interval);
1897 			q->avg_peak_bandwidth =
1898 				cake_ewma(q->avg_peak_bandwidth, b,
1899 					  b > q->avg_peak_bandwidth ? 2 : 8);
1900 			q->avg_window_bytes = 0;
1901 			q->avg_window_begin = now;
1902 
1903 			if (ktime_after(now,
1904 					ktime_add_ms(q->last_reconfig_time,
1905 						     250))) {
1906 				q->config->rate_bps = (q->avg_peak_bandwidth * 15) >> 4;
1907 				cake_reconfigure(sch);
1908 			}
1909 		}
1910 	} else {
1911 		q->avg_window_bytes = 0;
1912 		q->last_packet_time = now;
1913 	}
1914 
1915 	/* flowchain */
1916 	if (!flow->set || flow->set == CAKE_SET_DECAYING) {
1917 		if (!flow->set) {
1918 			list_add_tail(&flow->flowchain, &b->new_flows);
1919 		} else {
1920 			b->decaying_flow_count--;
1921 			list_move_tail(&flow->flowchain, &b->new_flows);
1922 		}
1923 		flow->set = CAKE_SET_SPARSE;
1924 		b->sparse_flow_count++;
1925 
1926 		flow->deficit = cake_get_flow_quantum(b, flow, q->config->flow_mode);
1927 	} else if (flow->set == CAKE_SET_SPARSE_WAIT) {
1928 		/* this flow was empty, accounted as a sparse flow, but actually
1929 		 * in the bulk rotation.
1930 		 */
1931 		flow->set = CAKE_SET_BULK;
1932 		b->sparse_flow_count--;
1933 		b->bulk_flow_count++;
1934 
1935 		cake_inc_srchost_bulk_flow_count(b, flow, q->config->flow_mode);
1936 		cake_inc_dsthost_bulk_flow_count(b, flow, q->config->flow_mode);
1937 	}
1938 
1939 	if (q->buffer_used > q->buffer_max_used)
1940 		q->buffer_max_used = q->buffer_used;
1941 
1942 	if (q->buffer_used <= q->buffer_limit)
1943 		return NET_XMIT_SUCCESS;
1944 
1945 	prev_qlen = sch->q.qlen;
1946 	prev_backlog = sch->qstats.backlog;
1947 
1948 	while (q->buffer_used > q->buffer_limit) {
1949 		drop_id = cake_drop(sch, to_free);
1950 		if ((drop_id >> 16) == tin &&
1951 		    (drop_id & 0xFFFF) == idx)
1952 			same_flow = true;
1953 	}
1954 
1955 	prev_qlen -= sch->q.qlen;
1956 	prev_backlog -= sch->qstats.backlog;
1957 	b->drop_overlimit += prev_qlen;
1958 
1959 	if (same_flow) {
1960 		qdisc_tree_reduce_backlog(sch, prev_qlen - 1,
1961 					  prev_backlog - len);
1962 		return NET_XMIT_CN;
1963 	}
1964 	qdisc_tree_reduce_backlog(sch, prev_qlen, prev_backlog);
1965 	return NET_XMIT_SUCCESS;
1966 }
1967 
1968 static struct sk_buff *cake_dequeue_one(struct Qdisc *sch)
1969 {
1970 	struct cake_sched_data *q = qdisc_priv(sch);
1971 	struct cake_tin_data *b = &q->tins[q->cur_tin];
1972 	struct cake_flow *flow = &b->flows[q->cur_flow];
1973 	struct sk_buff *skb = NULL;
1974 	u32 len;
1975 
1976 	if (flow->head) {
1977 		skb = dequeue_head(flow);
1978 		len = qdisc_pkt_len(skb);
1979 		b->backlogs[q->cur_flow] -= len;
1980 		b->tin_backlog		 -= len;
1981 		sch->qstats.backlog      -= len;
1982 		q->buffer_used		 -= skb->truesize;
1983 		sch->q.qlen--;
1984 
1985 		if (q->overflow_timeout)
1986 			cake_heapify(q, b->overflow_idx[q->cur_flow]);
1987 	}
1988 	return skb;
1989 }
1990 
1991 /* Discard leftover packets from a tin no longer in use. */
1992 static void cake_clear_tin(struct Qdisc *sch, u16 tin)
1993 {
1994 	struct cake_sched_data *q = qdisc_priv(sch);
1995 	struct sk_buff *skb;
1996 
1997 	q->cur_tin = tin;
1998 	for (q->cur_flow = 0; q->cur_flow < CAKE_QUEUES; q->cur_flow++)
1999 		while (!!(skb = cake_dequeue_one(sch)))
2000 			kfree_skb_reason(skb, SKB_DROP_REASON_QUEUE_PURGE);
2001 }
2002 
2003 static struct sk_buff *cake_dequeue(struct Qdisc *sch)
2004 {
2005 	struct cake_sched_data *q = qdisc_priv(sch);
2006 	struct cake_tin_data *b = &q->tins[q->cur_tin];
2007 	enum skb_drop_reason reason;
2008 	ktime_t now = ktime_get();
2009 	struct cake_flow *flow;
2010 	struct list_head *head;
2011 	bool first_flow = true;
2012 	struct sk_buff *skb;
2013 	u64 delay;
2014 	u32 len;
2015 
2016 	if (q->config->is_shared && now - q->last_checked_active >= q->config->sync_time) {
2017 		struct net_device *dev = qdisc_dev(sch);
2018 		struct cake_sched_data *other_priv;
2019 		u64 new_rate = q->config->rate_bps;
2020 		u64 other_qlen, other_last_active;
2021 		struct Qdisc *other_sch;
2022 		u32 num_active_qs = 1;
2023 		unsigned int ntx;
2024 
2025 		for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
2026 			other_sch = rcu_dereference(netdev_get_tx_queue(dev, ntx)->qdisc_sleeping);
2027 			other_priv = qdisc_priv(other_sch);
2028 
2029 			if (other_priv == q)
2030 				continue;
2031 
2032 			other_qlen = READ_ONCE(other_sch->q.qlen);
2033 			other_last_active = READ_ONCE(other_priv->last_active);
2034 
2035 			if (other_qlen || other_last_active > q->last_checked_active)
2036 				num_active_qs++;
2037 		}
2038 
2039 		if (num_active_qs > 1)
2040 			new_rate = div64_u64(q->config->rate_bps, num_active_qs);
2041 
2042 		/* mtu = 0 is used to only update the rate and not mess with cobalt params */
2043 		cake_set_rate(b, new_rate, 0, 0, 0);
2044 		q->last_checked_active = now;
2045 		q->active_queues = num_active_qs;
2046 		q->rate_ns = b->tin_rate_ns;
2047 		q->rate_shft = b->tin_rate_shft;
2048 	}
2049 
2050 begin:
2051 	if (!sch->q.qlen)
2052 		return NULL;
2053 
2054 	/* global hard shaper */
2055 	if (ktime_after(q->time_next_packet, now) &&
2056 	    ktime_after(q->failsafe_next_packet, now)) {
2057 		u64 next = min(ktime_to_ns(q->time_next_packet),
2058 			       ktime_to_ns(q->failsafe_next_packet));
2059 
2060 		sch->qstats.overlimits++;
2061 		qdisc_watchdog_schedule_ns(&q->watchdog, next);
2062 		return NULL;
2063 	}
2064 
2065 	/* Choose a class to work on. */
2066 	if (!q->rate_ns) {
2067 		/* In unlimited mode, can't rely on shaper timings, just balance
2068 		 * with DRR
2069 		 */
2070 		bool wrapped = false, empty = true;
2071 
2072 		while (b->tin_deficit < 0 ||
2073 		       !(b->sparse_flow_count + b->bulk_flow_count)) {
2074 			if (b->tin_deficit <= 0)
2075 				b->tin_deficit += b->tin_quantum;
2076 			if (b->sparse_flow_count + b->bulk_flow_count)
2077 				empty = false;
2078 
2079 			q->cur_tin++;
2080 			b++;
2081 			if (q->cur_tin >= q->tin_cnt) {
2082 				q->cur_tin = 0;
2083 				b = q->tins;
2084 
2085 				if (wrapped) {
2086 					/* It's possible for q->qlen to be
2087 					 * nonzero when we actually have no
2088 					 * packets anywhere.
2089 					 */
2090 					if (empty)
2091 						return NULL;
2092 				} else {
2093 					wrapped = true;
2094 				}
2095 			}
2096 		}
2097 	} else {
2098 		/* In shaped mode, choose:
2099 		 * - Highest-priority tin with queue and meeting schedule, or
2100 		 * - The earliest-scheduled tin with queue.
2101 		 */
2102 		ktime_t best_time = KTIME_MAX;
2103 		int tin, best_tin = 0;
2104 
2105 		for (tin = 0; tin < q->tin_cnt; tin++) {
2106 			b = q->tins + tin;
2107 			if ((b->sparse_flow_count + b->bulk_flow_count) > 0) {
2108 				ktime_t time_to_pkt = \
2109 					ktime_sub(b->time_next_packet, now);
2110 
2111 				if (ktime_to_ns(time_to_pkt) <= 0 ||
2112 				    ktime_compare(time_to_pkt,
2113 						  best_time) <= 0) {
2114 					best_time = time_to_pkt;
2115 					best_tin = tin;
2116 				}
2117 			}
2118 		}
2119 
2120 		q->cur_tin = best_tin;
2121 		b = q->tins + best_tin;
2122 
2123 		/* No point in going further if no packets to deliver. */
2124 		if (unlikely(!(b->sparse_flow_count + b->bulk_flow_count)))
2125 			return NULL;
2126 	}
2127 
2128 retry:
2129 	/* service this class */
2130 	head = &b->decaying_flows;
2131 	if (!first_flow || list_empty(head)) {
2132 		head = &b->new_flows;
2133 		if (list_empty(head)) {
2134 			head = &b->old_flows;
2135 			if (unlikely(list_empty(head))) {
2136 				head = &b->decaying_flows;
2137 				if (unlikely(list_empty(head)))
2138 					goto begin;
2139 			}
2140 		}
2141 	}
2142 	flow = list_first_entry(head, struct cake_flow, flowchain);
2143 	q->cur_flow = flow - b->flows;
2144 	first_flow = false;
2145 
2146 	/* flow isolation (DRR++) */
2147 	if (flow->deficit <= 0) {
2148 		/* Keep all flows with deficits out of the sparse and decaying
2149 		 * rotations.  No non-empty flow can go into the decaying
2150 		 * rotation, so they can't get deficits
2151 		 */
2152 		if (flow->set == CAKE_SET_SPARSE) {
2153 			if (flow->head) {
2154 				b->sparse_flow_count--;
2155 				b->bulk_flow_count++;
2156 
2157 				cake_inc_srchost_bulk_flow_count(b, flow, q->config->flow_mode);
2158 				cake_inc_dsthost_bulk_flow_count(b, flow, q->config->flow_mode);
2159 
2160 				flow->set = CAKE_SET_BULK;
2161 			} else {
2162 				/* we've moved it to the bulk rotation for
2163 				 * correct deficit accounting but we still want
2164 				 * to count it as a sparse flow, not a bulk one.
2165 				 */
2166 				flow->set = CAKE_SET_SPARSE_WAIT;
2167 			}
2168 		}
2169 
2170 		flow->deficit += cake_get_flow_quantum(b, flow, q->config->flow_mode);
2171 		list_move_tail(&flow->flowchain, &b->old_flows);
2172 
2173 		goto retry;
2174 	}
2175 
2176 	/* Retrieve a packet via the AQM */
2177 	while (1) {
2178 		skb = cake_dequeue_one(sch);
2179 		if (!skb) {
2180 			/* this queue was actually empty */
2181 			if (cobalt_queue_empty(&flow->cvars, &b->cparams, now))
2182 				b->unresponsive_flow_count--;
2183 
2184 			if (flow->cvars.p_drop || flow->cvars.count ||
2185 			    ktime_before(now, flow->cvars.drop_next)) {
2186 				/* keep in the flowchain until the state has
2187 				 * decayed to rest
2188 				 */
2189 				list_move_tail(&flow->flowchain,
2190 					       &b->decaying_flows);
2191 				if (flow->set == CAKE_SET_BULK) {
2192 					b->bulk_flow_count--;
2193 
2194 					cake_dec_srchost_bulk_flow_count(b, flow, q->config->flow_mode);
2195 					cake_dec_dsthost_bulk_flow_count(b, flow, q->config->flow_mode);
2196 
2197 					b->decaying_flow_count++;
2198 				} else if (flow->set == CAKE_SET_SPARSE ||
2199 					   flow->set == CAKE_SET_SPARSE_WAIT) {
2200 					b->sparse_flow_count--;
2201 					b->decaying_flow_count++;
2202 				}
2203 				flow->set = CAKE_SET_DECAYING;
2204 			} else {
2205 				/* remove empty queue from the flowchain */
2206 				list_del_init(&flow->flowchain);
2207 				if (flow->set == CAKE_SET_SPARSE ||
2208 				    flow->set == CAKE_SET_SPARSE_WAIT)
2209 					b->sparse_flow_count--;
2210 				else if (flow->set == CAKE_SET_BULK) {
2211 					b->bulk_flow_count--;
2212 
2213 					cake_dec_srchost_bulk_flow_count(b, flow, q->config->flow_mode);
2214 					cake_dec_dsthost_bulk_flow_count(b, flow, q->config->flow_mode);
2215 				} else
2216 					b->decaying_flow_count--;
2217 
2218 				flow->set = CAKE_SET_NONE;
2219 			}
2220 			goto begin;
2221 		}
2222 
2223 		reason = cobalt_should_drop(&flow->cvars, &b->cparams, now, skb,
2224 					    (b->bulk_flow_count *
2225 					     !!(q->config->rate_flags &
2226 						CAKE_FLAG_INGRESS)));
2227 		/* Last packet in queue may be marked, shouldn't be dropped */
2228 		if (reason == SKB_NOT_DROPPED_YET || !flow->head)
2229 			break;
2230 
2231 		/* drop this packet, get another one */
2232 		if (q->config->rate_flags & CAKE_FLAG_INGRESS) {
2233 			len = cake_advance_shaper(q, b, skb,
2234 						  now, true);
2235 			flow->deficit -= len;
2236 			b->tin_deficit -= len;
2237 		}
2238 		flow->dropped++;
2239 		b->tin_dropped++;
2240 		qdisc_tree_reduce_backlog(sch, 1, qdisc_pkt_len(skb));
2241 		qdisc_qstats_drop(sch);
2242 		qdisc_dequeue_drop(sch, skb, reason);
2243 		if (q->config->rate_flags & CAKE_FLAG_INGRESS)
2244 			goto retry;
2245 	}
2246 
2247 	b->tin_ecn_mark += !!flow->cvars.ecn_marked;
2248 	qdisc_bstats_update(sch, skb);
2249 	WRITE_ONCE(q->last_active, now);
2250 
2251 	/* collect delay stats */
2252 	delay = ktime_to_ns(ktime_sub(now, cobalt_get_enqueue_time(skb)));
2253 	b->avge_delay = cake_ewma(b->avge_delay, delay, 8);
2254 	b->peak_delay = cake_ewma(b->peak_delay, delay,
2255 				  delay > b->peak_delay ? 2 : 8);
2256 	b->base_delay = cake_ewma(b->base_delay, delay,
2257 				  delay < b->base_delay ? 2 : 8);
2258 
2259 	len = cake_advance_shaper(q, b, skb, now, false);
2260 	flow->deficit -= len;
2261 	b->tin_deficit -= len;
2262 
2263 	if (ktime_after(q->time_next_packet, now) && sch->q.qlen) {
2264 		u64 next = min(ktime_to_ns(q->time_next_packet),
2265 			       ktime_to_ns(q->failsafe_next_packet));
2266 
2267 		qdisc_watchdog_schedule_ns(&q->watchdog, next);
2268 	} else if (!sch->q.qlen) {
2269 		int i;
2270 
2271 		for (i = 0; i < q->tin_cnt; i++) {
2272 			if (q->tins[i].decaying_flow_count) {
2273 				ktime_t next = \
2274 					ktime_add_ns(now,
2275 						     q->tins[i].cparams.target);
2276 
2277 				qdisc_watchdog_schedule_ns(&q->watchdog,
2278 							   ktime_to_ns(next));
2279 				break;
2280 			}
2281 		}
2282 	}
2283 
2284 	if (q->overflow_timeout)
2285 		q->overflow_timeout--;
2286 
2287 	return skb;
2288 }
2289 
2290 static void cake_reset(struct Qdisc *sch)
2291 {
2292 	struct cake_sched_data *q = qdisc_priv(sch);
2293 	u32 c;
2294 
2295 	if (!q->tins)
2296 		return;
2297 
2298 	for (c = 0; c < CAKE_MAX_TINS; c++)
2299 		cake_clear_tin(sch, c);
2300 }
2301 
2302 static const struct nla_policy cake_policy[TCA_CAKE_MAX + 1] = {
2303 	[TCA_CAKE_BASE_RATE64]   = { .type = NLA_U64 },
2304 	[TCA_CAKE_DIFFSERV_MODE] = { .type = NLA_U32 },
2305 	[TCA_CAKE_ATM]		 = { .type = NLA_U32 },
2306 	[TCA_CAKE_FLOW_MODE]     = { .type = NLA_U32 },
2307 	[TCA_CAKE_OVERHEAD]      = { .type = NLA_S32 },
2308 	[TCA_CAKE_RTT]		 = { .type = NLA_U32 },
2309 	[TCA_CAKE_TARGET]	 = { .type = NLA_U32 },
2310 	[TCA_CAKE_AUTORATE]      = { .type = NLA_U32 },
2311 	[TCA_CAKE_MEMORY]	 = { .type = NLA_U32 },
2312 	[TCA_CAKE_NAT]		 = { .type = NLA_U32 },
2313 	[TCA_CAKE_RAW]		 = { .type = NLA_U32 },
2314 	[TCA_CAKE_WASH]		 = { .type = NLA_U32 },
2315 	[TCA_CAKE_MPU]		 = { .type = NLA_U32 },
2316 	[TCA_CAKE_INGRESS]	 = { .type = NLA_U32 },
2317 	[TCA_CAKE_ACK_FILTER]	 = { .type = NLA_U32 },
2318 	[TCA_CAKE_SPLIT_GSO]	 = { .type = NLA_U32 },
2319 	[TCA_CAKE_FWMARK]	 = { .type = NLA_U32 },
2320 };
2321 
2322 static void cake_set_rate(struct cake_tin_data *b, u64 rate, u32 mtu,
2323 			  u64 target_ns, u64 rtt_est_ns)
2324 {
2325 	/* convert byte-rate into time-per-byte
2326 	 * so it will always unwedge in reasonable time.
2327 	 */
2328 	static const u64 MIN_RATE = 64;
2329 	u32 byte_target = mtu;
2330 	u64 byte_target_ns;
2331 	u8  rate_shft = 0;
2332 	u64 rate_ns = 0;
2333 
2334 	b->flow_quantum = 1514;
2335 	if (rate) {
2336 		b->flow_quantum = max(min(rate >> 12, 1514ULL), 300ULL);
2337 		rate_shft = 34;
2338 		rate_ns = ((u64)NSEC_PER_SEC) << rate_shft;
2339 		rate_ns = div64_u64(rate_ns, max(MIN_RATE, rate));
2340 		while (!!(rate_ns >> 34)) {
2341 			rate_ns >>= 1;
2342 			rate_shft--;
2343 		}
2344 	} /* else unlimited, ie. zero delay */
2345 
2346 	b->tin_rate_bps  = rate;
2347 	b->tin_rate_ns   = rate_ns;
2348 	b->tin_rate_shft = rate_shft;
2349 
2350 	if (mtu == 0)
2351 		return;
2352 
2353 	byte_target_ns = (byte_target * rate_ns) >> rate_shft;
2354 
2355 	b->cparams.target = max((byte_target_ns * 3) / 2, target_ns);
2356 	b->cparams.interval = max(rtt_est_ns +
2357 				     b->cparams.target - target_ns,
2358 				     b->cparams.target * 2);
2359 	b->cparams.mtu_time = byte_target_ns;
2360 	b->cparams.p_inc = 1 << 24; /* 1/256 */
2361 	b->cparams.p_dec = 1 << 20; /* 1/4096 */
2362 }
2363 
2364 static int cake_config_besteffort(struct Qdisc *sch)
2365 {
2366 	struct cake_sched_data *q = qdisc_priv(sch);
2367 	struct cake_tin_data *b = &q->tins[0];
2368 	u32 mtu = psched_mtu(qdisc_dev(sch));
2369 	u64 rate = q->config->rate_bps;
2370 
2371 	q->tin_cnt = 1;
2372 
2373 	q->tin_index = besteffort;
2374 	q->tin_order = normal_order;
2375 
2376 	cake_set_rate(b, rate, mtu,
2377 		      us_to_ns(q->config->target), us_to_ns(q->config->interval));
2378 	b->tin_quantum = 65535;
2379 
2380 	return 0;
2381 }
2382 
2383 static int cake_config_precedence(struct Qdisc *sch)
2384 {
2385 	/* convert high-level (user visible) parameters into internal format */
2386 	struct cake_sched_data *q = qdisc_priv(sch);
2387 	u32 mtu = psched_mtu(qdisc_dev(sch));
2388 	u64 rate = q->config->rate_bps;
2389 	u32 quantum = 256;
2390 	u32 i;
2391 
2392 	q->tin_cnt = 8;
2393 	q->tin_index = precedence;
2394 	q->tin_order = normal_order;
2395 
2396 	for (i = 0; i < q->tin_cnt; i++) {
2397 		struct cake_tin_data *b = &q->tins[i];
2398 
2399 		cake_set_rate(b, rate, mtu, us_to_ns(q->config->target),
2400 			      us_to_ns(q->config->interval));
2401 
2402 		b->tin_quantum = max_t(u16, 1U, quantum);
2403 
2404 		/* calculate next class's parameters */
2405 		rate  *= 7;
2406 		rate >>= 3;
2407 
2408 		quantum  *= 7;
2409 		quantum >>= 3;
2410 	}
2411 
2412 	return 0;
2413 }
2414 
2415 /*	List of known Diffserv codepoints:
2416  *
2417  *	Default Forwarding (DF/CS0) - Best Effort
2418  *	Max Throughput (TOS2)
2419  *	Min Delay (TOS4)
2420  *	LLT "La" (TOS5)
2421  *	Assured Forwarding 1 (AF1x) - x3
2422  *	Assured Forwarding 2 (AF2x) - x3
2423  *	Assured Forwarding 3 (AF3x) - x3
2424  *	Assured Forwarding 4 (AF4x) - x3
2425  *	Precedence Class 1 (CS1)
2426  *	Precedence Class 2 (CS2)
2427  *	Precedence Class 3 (CS3)
2428  *	Precedence Class 4 (CS4)
2429  *	Precedence Class 5 (CS5)
2430  *	Precedence Class 6 (CS6)
2431  *	Precedence Class 7 (CS7)
2432  *	Voice Admit (VA)
2433  *	Expedited Forwarding (EF)
2434  *	Lower Effort (LE)
2435  *
2436  *	Total 26 codepoints.
2437  */
2438 
2439 /*	List of traffic classes in RFC 4594, updated by RFC 8622:
2440  *		(roughly descending order of contended priority)
2441  *		(roughly ascending order of uncontended throughput)
2442  *
2443  *	Network Control (CS6,CS7)      - routing traffic
2444  *	Telephony (EF,VA)         - aka. VoIP streams
2445  *	Signalling (CS5)               - VoIP setup
2446  *	Multimedia Conferencing (AF4x) - aka. video calls
2447  *	Realtime Interactive (CS4)     - eg. games
2448  *	Multimedia Streaming (AF3x)    - eg. YouTube, NetFlix, Twitch
2449  *	Broadcast Video (CS3)
2450  *	Low-Latency Data (AF2x,TOS4)      - eg. database
2451  *	Ops, Admin, Management (CS2)      - eg. ssh
2452  *	Standard Service (DF & unrecognised codepoints)
2453  *	High-Throughput Data (AF1x,TOS2)  - eg. web traffic
2454  *	Low-Priority Data (LE,CS1)        - eg. BitTorrent
2455  *
2456  *	Total 12 traffic classes.
2457  */
2458 
2459 static int cake_config_diffserv8(struct Qdisc *sch)
2460 {
2461 /*	Pruned list of traffic classes for typical applications:
2462  *
2463  *		Network Control          (CS6, CS7)
2464  *		Minimum Latency          (EF, VA, CS5, CS4)
2465  *		Interactive Shell        (CS2)
2466  *		Low Latency Transactions (AF2x, TOS4)
2467  *		Video Streaming          (AF4x, AF3x, CS3)
2468  *		Bog Standard             (DF etc.)
2469  *		High Throughput          (AF1x, TOS2, CS1)
2470  *		Background Traffic       (LE)
2471  *
2472  *		Total 8 traffic classes.
2473  */
2474 
2475 	struct cake_sched_data *q = qdisc_priv(sch);
2476 	u32 mtu = psched_mtu(qdisc_dev(sch));
2477 	u64 rate = q->config->rate_bps;
2478 	u32 quantum = 256;
2479 	u32 i;
2480 
2481 	q->tin_cnt = 8;
2482 
2483 	/* codepoint to class mapping */
2484 	q->tin_index = diffserv8;
2485 	q->tin_order = normal_order;
2486 
2487 	/* class characteristics */
2488 	for (i = 0; i < q->tin_cnt; i++) {
2489 		struct cake_tin_data *b = &q->tins[i];
2490 
2491 		cake_set_rate(b, rate, mtu, us_to_ns(q->config->target),
2492 			      us_to_ns(q->config->interval));
2493 
2494 		b->tin_quantum = max_t(u16, 1U, quantum);
2495 
2496 		/* calculate next class's parameters */
2497 		rate  *= 7;
2498 		rate >>= 3;
2499 
2500 		quantum  *= 7;
2501 		quantum >>= 3;
2502 	}
2503 
2504 	return 0;
2505 }
2506 
2507 static int cake_config_diffserv4(struct Qdisc *sch)
2508 {
2509 /*  Further pruned list of traffic classes for four-class system:
2510  *
2511  *	    Latency Sensitive  (CS7, CS6, EF, VA, CS5, CS4)
2512  *	    Streaming Media    (AF4x, AF3x, CS3, AF2x, TOS4, CS2)
2513  *	    Best Effort        (DF, AF1x, TOS2, and those not specified)
2514  *	    Background Traffic (LE, CS1)
2515  *
2516  *		Total 4 traffic classes.
2517  */
2518 
2519 	struct cake_sched_data *q = qdisc_priv(sch);
2520 	u32 mtu = psched_mtu(qdisc_dev(sch));
2521 	u64 rate = q->config->rate_bps;
2522 	u32 quantum = 1024;
2523 
2524 	q->tin_cnt = 4;
2525 
2526 	/* codepoint to class mapping */
2527 	q->tin_index = diffserv4;
2528 	q->tin_order = bulk_order;
2529 
2530 	/* class characteristics */
2531 	cake_set_rate(&q->tins[0], rate, mtu,
2532 		      us_to_ns(q->config->target), us_to_ns(q->config->interval));
2533 	cake_set_rate(&q->tins[1], rate >> 4, mtu,
2534 		      us_to_ns(q->config->target), us_to_ns(q->config->interval));
2535 	cake_set_rate(&q->tins[2], rate >> 1, mtu,
2536 		      us_to_ns(q->config->target), us_to_ns(q->config->interval));
2537 	cake_set_rate(&q->tins[3], rate >> 2, mtu,
2538 		      us_to_ns(q->config->target), us_to_ns(q->config->interval));
2539 
2540 	/* bandwidth-sharing weights */
2541 	q->tins[0].tin_quantum = quantum;
2542 	q->tins[1].tin_quantum = quantum >> 4;
2543 	q->tins[2].tin_quantum = quantum >> 1;
2544 	q->tins[3].tin_quantum = quantum >> 2;
2545 
2546 	return 0;
2547 }
2548 
2549 static int cake_config_diffserv3(struct Qdisc *sch)
2550 {
2551 /*  Simplified Diffserv structure with 3 tins.
2552  *		Latency Sensitive	(CS7, CS6, EF, VA, TOS4)
2553  *		Best Effort
2554  *		Low Priority		(LE, CS1)
2555  */
2556 	struct cake_sched_data *q = qdisc_priv(sch);
2557 	u32 mtu = psched_mtu(qdisc_dev(sch));
2558 	u64 rate = q->config->rate_bps;
2559 	u32 quantum = 1024;
2560 
2561 	q->tin_cnt = 3;
2562 
2563 	/* codepoint to class mapping */
2564 	q->tin_index = diffserv3;
2565 	q->tin_order = bulk_order;
2566 
2567 	/* class characteristics */
2568 	cake_set_rate(&q->tins[0], rate, mtu,
2569 		      us_to_ns(q->config->target), us_to_ns(q->config->interval));
2570 	cake_set_rate(&q->tins[1], rate >> 4, mtu,
2571 		      us_to_ns(q->config->target), us_to_ns(q->config->interval));
2572 	cake_set_rate(&q->tins[2], rate >> 2, mtu,
2573 		      us_to_ns(q->config->target), us_to_ns(q->config->interval));
2574 
2575 	/* bandwidth-sharing weights */
2576 	q->tins[0].tin_quantum = quantum;
2577 	q->tins[1].tin_quantum = quantum >> 4;
2578 	q->tins[2].tin_quantum = quantum >> 2;
2579 
2580 	return 0;
2581 }
2582 
2583 static void cake_reconfigure(struct Qdisc *sch)
2584 {
2585 	struct cake_sched_data *qd = qdisc_priv(sch);
2586 	struct cake_sched_config *q = qd->config;
2587 	int c, ft;
2588 
2589 	switch (q->tin_mode) {
2590 	case CAKE_DIFFSERV_BESTEFFORT:
2591 		ft = cake_config_besteffort(sch);
2592 		break;
2593 
2594 	case CAKE_DIFFSERV_PRECEDENCE:
2595 		ft = cake_config_precedence(sch);
2596 		break;
2597 
2598 	case CAKE_DIFFSERV_DIFFSERV8:
2599 		ft = cake_config_diffserv8(sch);
2600 		break;
2601 
2602 	case CAKE_DIFFSERV_DIFFSERV4:
2603 		ft = cake_config_diffserv4(sch);
2604 		break;
2605 
2606 	case CAKE_DIFFSERV_DIFFSERV3:
2607 	default:
2608 		ft = cake_config_diffserv3(sch);
2609 		break;
2610 	}
2611 
2612 	for (c = qd->tin_cnt; c < CAKE_MAX_TINS; c++) {
2613 		cake_clear_tin(sch, c);
2614 		qd->tins[c].cparams.mtu_time = qd->tins[ft].cparams.mtu_time;
2615 	}
2616 
2617 	qd->rate_ns   = qd->tins[ft].tin_rate_ns;
2618 	qd->rate_shft = qd->tins[ft].tin_rate_shft;
2619 
2620 	if (q->buffer_config_limit) {
2621 		qd->buffer_limit = q->buffer_config_limit;
2622 	} else if (q->rate_bps) {
2623 		u64 t = q->rate_bps * q->interval;
2624 
2625 		do_div(t, USEC_PER_SEC / 4);
2626 		qd->buffer_limit = max_t(u32, t, 4U << 20);
2627 	} else {
2628 		qd->buffer_limit = ~0;
2629 	}
2630 
2631 	sch->flags &= ~TCQ_F_CAN_BYPASS;
2632 
2633 	qd->buffer_limit = min(qd->buffer_limit,
2634 			       max(sch->limit * psched_mtu(qdisc_dev(sch)),
2635 				   q->buffer_config_limit));
2636 }
2637 
2638 static int cake_config_change(struct cake_sched_config *q, struct nlattr *opt,
2639 			      struct netlink_ext_ack *extack, bool *overhead_changed)
2640 {
2641 	struct nlattr *tb[TCA_CAKE_MAX + 1];
2642 	u16 rate_flags = q->rate_flags;
2643 	u8 flow_mode = q->flow_mode;
2644 	int err;
2645 
2646 	err = nla_parse_nested_deprecated(tb, TCA_CAKE_MAX, opt, cake_policy,
2647 					  extack);
2648 	if (err < 0)
2649 		return err;
2650 
2651 	if (tb[TCA_CAKE_NAT]) {
2652 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
2653 		flow_mode &= ~CAKE_FLOW_NAT_FLAG;
2654 		flow_mode |= CAKE_FLOW_NAT_FLAG *
2655 			!!nla_get_u32(tb[TCA_CAKE_NAT]);
2656 #else
2657 		NL_SET_ERR_MSG_ATTR(extack, tb[TCA_CAKE_NAT],
2658 				    "No conntrack support in kernel");
2659 		return -EOPNOTSUPP;
2660 #endif
2661 	}
2662 
2663 	if (tb[TCA_CAKE_AUTORATE]) {
2664 		if (!!nla_get_u32(tb[TCA_CAKE_AUTORATE])) {
2665 			if (q->is_shared) {
2666 				NL_SET_ERR_MSG_ATTR(extack, tb[TCA_CAKE_AUTORATE],
2667 						    "Can't use autorate-ingress with cake_mq");
2668 				return -EOPNOTSUPP;
2669 			}
2670 			rate_flags |= CAKE_FLAG_AUTORATE_INGRESS;
2671 		} else {
2672 			rate_flags &= ~CAKE_FLAG_AUTORATE_INGRESS;
2673 		}
2674 	}
2675 
2676 	if (tb[TCA_CAKE_BASE_RATE64])
2677 		WRITE_ONCE(q->rate_bps,
2678 			   nla_get_u64(tb[TCA_CAKE_BASE_RATE64]));
2679 
2680 	if (tb[TCA_CAKE_DIFFSERV_MODE])
2681 		WRITE_ONCE(q->tin_mode,
2682 			   nla_get_u32(tb[TCA_CAKE_DIFFSERV_MODE]));
2683 
2684 	if (tb[TCA_CAKE_WASH]) {
2685 		if (!!nla_get_u32(tb[TCA_CAKE_WASH]))
2686 			rate_flags |= CAKE_FLAG_WASH;
2687 		else
2688 			rate_flags &= ~CAKE_FLAG_WASH;
2689 	}
2690 
2691 	if (tb[TCA_CAKE_FLOW_MODE])
2692 		flow_mode = ((flow_mode & CAKE_FLOW_NAT_FLAG) |
2693 				(nla_get_u32(tb[TCA_CAKE_FLOW_MODE]) &
2694 					CAKE_FLOW_MASK));
2695 
2696 	if (tb[TCA_CAKE_ATM])
2697 		WRITE_ONCE(q->atm_mode,
2698 			   nla_get_u32(tb[TCA_CAKE_ATM]));
2699 
2700 	if (tb[TCA_CAKE_OVERHEAD]) {
2701 		WRITE_ONCE(q->rate_overhead,
2702 			   nla_get_s32(tb[TCA_CAKE_OVERHEAD]));
2703 		rate_flags |= CAKE_FLAG_OVERHEAD;
2704 		*overhead_changed = true;
2705 	}
2706 
2707 	if (tb[TCA_CAKE_RAW]) {
2708 		rate_flags &= ~CAKE_FLAG_OVERHEAD;
2709 		*overhead_changed = true;
2710 	}
2711 
2712 	if (tb[TCA_CAKE_MPU])
2713 		WRITE_ONCE(q->rate_mpu,
2714 			   nla_get_u32(tb[TCA_CAKE_MPU]));
2715 
2716 	if (tb[TCA_CAKE_RTT]) {
2717 		u32 interval = nla_get_u32(tb[TCA_CAKE_RTT]);
2718 
2719 		WRITE_ONCE(q->interval, max(interval, 1U));
2720 	}
2721 
2722 	if (tb[TCA_CAKE_TARGET]) {
2723 		u32 target = nla_get_u32(tb[TCA_CAKE_TARGET]);
2724 
2725 		WRITE_ONCE(q->target, max(target, 1U));
2726 	}
2727 
2728 	if (tb[TCA_CAKE_INGRESS]) {
2729 		if (!!nla_get_u32(tb[TCA_CAKE_INGRESS]))
2730 			rate_flags |= CAKE_FLAG_INGRESS;
2731 		else
2732 			rate_flags &= ~CAKE_FLAG_INGRESS;
2733 	}
2734 
2735 	if (tb[TCA_CAKE_ACK_FILTER])
2736 		WRITE_ONCE(q->ack_filter,
2737 			   nla_get_u32(tb[TCA_CAKE_ACK_FILTER]));
2738 
2739 	if (tb[TCA_CAKE_MEMORY])
2740 		WRITE_ONCE(q->buffer_config_limit,
2741 			   nla_get_u32(tb[TCA_CAKE_MEMORY]));
2742 
2743 	if (tb[TCA_CAKE_SPLIT_GSO]) {
2744 		if (!!nla_get_u32(tb[TCA_CAKE_SPLIT_GSO]))
2745 			rate_flags |= CAKE_FLAG_SPLIT_GSO;
2746 		else
2747 			rate_flags &= ~CAKE_FLAG_SPLIT_GSO;
2748 	}
2749 
2750 	if (tb[TCA_CAKE_FWMARK]) {
2751 		WRITE_ONCE(q->fwmark_mask, nla_get_u32(tb[TCA_CAKE_FWMARK]));
2752 		WRITE_ONCE(q->fwmark_shft,
2753 			   q->fwmark_mask ? __ffs(q->fwmark_mask) : 0);
2754 	}
2755 
2756 	WRITE_ONCE(q->rate_flags, rate_flags);
2757 	WRITE_ONCE(q->flow_mode, flow_mode);
2758 
2759 	return 0;
2760 }
2761 
2762 static int cake_change(struct Qdisc *sch, struct nlattr *opt,
2763 		       struct netlink_ext_ack *extack)
2764 {
2765 	struct cake_sched_data *qd = qdisc_priv(sch);
2766 	struct cake_sched_config *q = qd->config;
2767 	bool overhead_changed = false;
2768 	int ret;
2769 
2770 	if (q->is_shared) {
2771 		NL_SET_ERR_MSG(extack, "can't reconfigure cake_mq sub-qdiscs");
2772 		return -EOPNOTSUPP;
2773 	}
2774 
2775 	ret = cake_config_change(q, opt, extack, &overhead_changed);
2776 	if (ret)
2777 		return ret;
2778 
2779 	if (overhead_changed) {
2780 		qd->max_netlen = 0;
2781 		qd->max_adjlen = 0;
2782 		qd->min_netlen = ~0;
2783 		qd->min_adjlen = ~0;
2784 	}
2785 
2786 	if (qd->tins) {
2787 		sch_tree_lock(sch);
2788 		cake_reconfigure(sch);
2789 		sch_tree_unlock(sch);
2790 	}
2791 
2792 	return 0;
2793 }
2794 
2795 static void cake_destroy(struct Qdisc *sch)
2796 {
2797 	struct cake_sched_data *q = qdisc_priv(sch);
2798 
2799 	qdisc_watchdog_cancel(&q->watchdog);
2800 	tcf_block_put(q->block);
2801 	kvfree(q->tins);
2802 }
2803 
2804 static void cake_config_init(struct cake_sched_config *q, bool is_shared)
2805 {
2806 	q->tin_mode = CAKE_DIFFSERV_DIFFSERV3;
2807 	q->flow_mode  = CAKE_FLOW_TRIPLE;
2808 
2809 	q->rate_bps = 0; /* unlimited by default */
2810 
2811 	q->interval = 100000; /* 100ms default */
2812 	q->target   =   5000; /* 5ms: codel RFC argues
2813 			       * for 5 to 10% of interval
2814 			       */
2815 	q->rate_flags |= CAKE_FLAG_SPLIT_GSO;
2816 	q->is_shared = is_shared;
2817 	q->sync_time = 200 * NSEC_PER_USEC;
2818 }
2819 
2820 static int cake_init(struct Qdisc *sch, struct nlattr *opt,
2821 		     struct netlink_ext_ack *extack)
2822 {
2823 	struct cake_sched_data *qd = qdisc_priv(sch);
2824 	struct cake_sched_config *q = &qd->initial_config;
2825 	int i, j, err;
2826 
2827 	cake_config_init(q, false);
2828 
2829 	sch->limit = 10240;
2830 	sch->flags |= TCQ_F_DEQUEUE_DROPS;
2831 
2832 	qd->cur_tin = 0;
2833 	qd->cur_flow  = 0;
2834 	qd->config = q;
2835 
2836 	qdisc_watchdog_init(&qd->watchdog, sch);
2837 
2838 	if (opt) {
2839 		err = cake_change(sch, opt, extack);
2840 		if (err)
2841 			return err;
2842 	}
2843 
2844 	err = tcf_block_get(&qd->block, &qd->filter_list, sch, extack);
2845 	if (err)
2846 		return err;
2847 
2848 	quantum_div[0] = ~0;
2849 	for (i = 1; i <= CAKE_QUEUES; i++)
2850 		quantum_div[i] = 65535 / i;
2851 
2852 	qd->tins = kvcalloc(CAKE_MAX_TINS, sizeof(struct cake_tin_data),
2853 			    GFP_KERNEL);
2854 	if (!qd->tins)
2855 		return -ENOMEM;
2856 
2857 	for (i = 0; i < CAKE_MAX_TINS; i++) {
2858 		struct cake_tin_data *b = qd->tins + i;
2859 
2860 		INIT_LIST_HEAD(&b->new_flows);
2861 		INIT_LIST_HEAD(&b->old_flows);
2862 		INIT_LIST_HEAD(&b->decaying_flows);
2863 		b->sparse_flow_count = 0;
2864 		b->bulk_flow_count = 0;
2865 		b->decaying_flow_count = 0;
2866 
2867 		for (j = 0; j < CAKE_QUEUES; j++) {
2868 			struct cake_flow *flow = b->flows + j;
2869 			u32 k = j * CAKE_MAX_TINS + i;
2870 
2871 			INIT_LIST_HEAD(&flow->flowchain);
2872 			cobalt_vars_init(&flow->cvars);
2873 
2874 			qd->overflow_heap[k].t = i;
2875 			qd->overflow_heap[k].b = j;
2876 			b->overflow_idx[j] = k;
2877 		}
2878 	}
2879 
2880 	cake_reconfigure(sch);
2881 	qd->avg_peak_bandwidth = q->rate_bps;
2882 	qd->min_netlen = ~0;
2883 	qd->min_adjlen = ~0;
2884 	qd->active_queues = 0;
2885 	qd->last_checked_active = 0;
2886 
2887 	return 0;
2888 }
2889 
2890 static void cake_config_replace(struct Qdisc *sch, struct cake_sched_config *cfg)
2891 {
2892 	struct cake_sched_data *qd = qdisc_priv(sch);
2893 
2894 	qd->config = cfg;
2895 	cake_reconfigure(sch);
2896 }
2897 
2898 static int cake_config_dump(struct cake_sched_config *q, struct sk_buff *skb)
2899 {
2900 	struct nlattr *opts;
2901 	u16 rate_flags;
2902 	u8 flow_mode;
2903 
2904 	opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
2905 	if (!opts)
2906 		goto nla_put_failure;
2907 
2908 	if (nla_put_u64_64bit(skb, TCA_CAKE_BASE_RATE64,
2909 			      READ_ONCE(q->rate_bps), TCA_CAKE_PAD))
2910 		goto nla_put_failure;
2911 
2912 	flow_mode = READ_ONCE(q->flow_mode);
2913 	if (nla_put_u32(skb, TCA_CAKE_FLOW_MODE, flow_mode & CAKE_FLOW_MASK))
2914 		goto nla_put_failure;
2915 
2916 	if (nla_put_u32(skb, TCA_CAKE_RTT, READ_ONCE(q->interval)))
2917 		goto nla_put_failure;
2918 
2919 	if (nla_put_u32(skb, TCA_CAKE_TARGET, READ_ONCE(q->target)))
2920 		goto nla_put_failure;
2921 
2922 	if (nla_put_u32(skb, TCA_CAKE_MEMORY,
2923 			READ_ONCE(q->buffer_config_limit)))
2924 		goto nla_put_failure;
2925 
2926 	rate_flags = READ_ONCE(q->rate_flags);
2927 	if (nla_put_u32(skb, TCA_CAKE_AUTORATE,
2928 			!!(rate_flags & CAKE_FLAG_AUTORATE_INGRESS)))
2929 		goto nla_put_failure;
2930 
2931 	if (nla_put_u32(skb, TCA_CAKE_INGRESS,
2932 			!!(rate_flags & CAKE_FLAG_INGRESS)))
2933 		goto nla_put_failure;
2934 
2935 	if (nla_put_u32(skb, TCA_CAKE_ACK_FILTER, READ_ONCE(q->ack_filter)))
2936 		goto nla_put_failure;
2937 
2938 	if (nla_put_u32(skb, TCA_CAKE_NAT,
2939 			!!(flow_mode & CAKE_FLOW_NAT_FLAG)))
2940 		goto nla_put_failure;
2941 
2942 	if (nla_put_u32(skb, TCA_CAKE_DIFFSERV_MODE, READ_ONCE(q->tin_mode)))
2943 		goto nla_put_failure;
2944 
2945 	if (nla_put_u32(skb, TCA_CAKE_WASH,
2946 			!!(rate_flags & CAKE_FLAG_WASH)))
2947 		goto nla_put_failure;
2948 
2949 	if (nla_put_u32(skb, TCA_CAKE_OVERHEAD, READ_ONCE(q->rate_overhead)))
2950 		goto nla_put_failure;
2951 
2952 	if (!(rate_flags & CAKE_FLAG_OVERHEAD))
2953 		if (nla_put_u32(skb, TCA_CAKE_RAW, 0))
2954 			goto nla_put_failure;
2955 
2956 	if (nla_put_u32(skb, TCA_CAKE_ATM, READ_ONCE(q->atm_mode)))
2957 		goto nla_put_failure;
2958 
2959 	if (nla_put_u32(skb, TCA_CAKE_MPU, READ_ONCE(q->rate_mpu)))
2960 		goto nla_put_failure;
2961 
2962 	if (nla_put_u32(skb, TCA_CAKE_SPLIT_GSO,
2963 			!!(rate_flags & CAKE_FLAG_SPLIT_GSO)))
2964 		goto nla_put_failure;
2965 
2966 	if (nla_put_u32(skb, TCA_CAKE_FWMARK, READ_ONCE(q->fwmark_mask)))
2967 		goto nla_put_failure;
2968 
2969 	return nla_nest_end(skb, opts);
2970 
2971 nla_put_failure:
2972 	return -1;
2973 }
2974 
2975 static int cake_dump(struct Qdisc *sch, struct sk_buff *skb)
2976 {
2977 	struct cake_sched_data *qd = qdisc_priv(sch);
2978 
2979 	return cake_config_dump(qd->config, skb);
2980 }
2981 
2982 static int cake_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
2983 {
2984 	struct nlattr *stats = nla_nest_start_noflag(d->skb, TCA_STATS_APP);
2985 	struct cake_sched_data *q = qdisc_priv(sch);
2986 	struct nlattr *tstats, *ts;
2987 	int i;
2988 
2989 	if (!stats)
2990 		return -1;
2991 
2992 #define PUT_STAT_U32(attr, data) do {				       \
2993 		if (nla_put_u32(d->skb, TCA_CAKE_STATS_ ## attr, data)) \
2994 			goto nla_put_failure;			       \
2995 	} while (0)
2996 #define PUT_STAT_U64(attr, data) do {				       \
2997 		if (nla_put_u64_64bit(d->skb, TCA_CAKE_STATS_ ## attr, \
2998 					data, TCA_CAKE_STATS_PAD)) \
2999 			goto nla_put_failure;			       \
3000 	} while (0)
3001 
3002 	PUT_STAT_U64(CAPACITY_ESTIMATE64, q->avg_peak_bandwidth);
3003 	PUT_STAT_U32(MEMORY_LIMIT, q->buffer_limit);
3004 	PUT_STAT_U32(MEMORY_USED, q->buffer_max_used);
3005 	PUT_STAT_U32(AVG_NETOFF, ((q->avg_netoff + 0x8000) >> 16));
3006 	PUT_STAT_U32(MAX_NETLEN, q->max_netlen);
3007 	PUT_STAT_U32(MAX_ADJLEN, q->max_adjlen);
3008 	PUT_STAT_U32(MIN_NETLEN, q->min_netlen);
3009 	PUT_STAT_U32(MIN_ADJLEN, q->min_adjlen);
3010 	PUT_STAT_U32(ACTIVE_QUEUES, q->active_queues);
3011 
3012 #undef PUT_STAT_U32
3013 #undef PUT_STAT_U64
3014 
3015 	tstats = nla_nest_start_noflag(d->skb, TCA_CAKE_STATS_TIN_STATS);
3016 	if (!tstats)
3017 		goto nla_put_failure;
3018 
3019 #define PUT_TSTAT_U32(attr, data) do {					\
3020 		if (nla_put_u32(d->skb, TCA_CAKE_TIN_STATS_ ## attr, data)) \
3021 			goto nla_put_failure;				\
3022 	} while (0)
3023 #define PUT_TSTAT_U64(attr, data) do {					\
3024 		if (nla_put_u64_64bit(d->skb, TCA_CAKE_TIN_STATS_ ## attr, \
3025 					data, TCA_CAKE_TIN_STATS_PAD))	\
3026 			goto nla_put_failure;				\
3027 	} while (0)
3028 
3029 	for (i = 0; i < q->tin_cnt; i++) {
3030 		struct cake_tin_data *b = &q->tins[q->tin_order[i]];
3031 
3032 		ts = nla_nest_start_noflag(d->skb, i + 1);
3033 		if (!ts)
3034 			goto nla_put_failure;
3035 
3036 		PUT_TSTAT_U64(THRESHOLD_RATE64, b->tin_rate_bps);
3037 		PUT_TSTAT_U64(SENT_BYTES64, b->bytes);
3038 		PUT_TSTAT_U32(BACKLOG_BYTES, b->tin_backlog);
3039 
3040 		PUT_TSTAT_U32(TARGET_US,
3041 			      ktime_to_us(ns_to_ktime(b->cparams.target)));
3042 		PUT_TSTAT_U32(INTERVAL_US,
3043 			      ktime_to_us(ns_to_ktime(b->cparams.interval)));
3044 
3045 		PUT_TSTAT_U32(SENT_PACKETS, b->packets);
3046 		PUT_TSTAT_U32(DROPPED_PACKETS, b->tin_dropped);
3047 		PUT_TSTAT_U32(ECN_MARKED_PACKETS, b->tin_ecn_mark);
3048 		PUT_TSTAT_U32(ACKS_DROPPED_PACKETS, b->ack_drops);
3049 
3050 		PUT_TSTAT_U32(PEAK_DELAY_US,
3051 			      ktime_to_us(ns_to_ktime(b->peak_delay)));
3052 		PUT_TSTAT_U32(AVG_DELAY_US,
3053 			      ktime_to_us(ns_to_ktime(b->avge_delay)));
3054 		PUT_TSTAT_U32(BASE_DELAY_US,
3055 			      ktime_to_us(ns_to_ktime(b->base_delay)));
3056 
3057 		PUT_TSTAT_U32(WAY_INDIRECT_HITS, b->way_hits);
3058 		PUT_TSTAT_U32(WAY_MISSES, b->way_misses);
3059 		PUT_TSTAT_U32(WAY_COLLISIONS, b->way_collisions);
3060 
3061 		PUT_TSTAT_U32(SPARSE_FLOWS, b->sparse_flow_count +
3062 					    b->decaying_flow_count);
3063 		PUT_TSTAT_U32(BULK_FLOWS, b->bulk_flow_count);
3064 		PUT_TSTAT_U32(UNRESPONSIVE_FLOWS, b->unresponsive_flow_count);
3065 		PUT_TSTAT_U32(MAX_SKBLEN, b->max_skblen);
3066 
3067 		PUT_TSTAT_U32(FLOW_QUANTUM, b->flow_quantum);
3068 		nla_nest_end(d->skb, ts);
3069 	}
3070 
3071 #undef PUT_TSTAT_U32
3072 #undef PUT_TSTAT_U64
3073 
3074 	nla_nest_end(d->skb, tstats);
3075 	return nla_nest_end(d->skb, stats);
3076 
3077 nla_put_failure:
3078 	nla_nest_cancel(d->skb, stats);
3079 	return -1;
3080 }
3081 
3082 static struct Qdisc *cake_leaf(struct Qdisc *sch, unsigned long arg)
3083 {
3084 	return NULL;
3085 }
3086 
3087 static unsigned long cake_find(struct Qdisc *sch, u32 classid)
3088 {
3089 	return 0;
3090 }
3091 
3092 static unsigned long cake_bind(struct Qdisc *sch, unsigned long parent,
3093 			       u32 classid)
3094 {
3095 	return 0;
3096 }
3097 
3098 static void cake_unbind(struct Qdisc *q, unsigned long cl)
3099 {
3100 }
3101 
3102 static struct tcf_block *cake_tcf_block(struct Qdisc *sch, unsigned long cl,
3103 					struct netlink_ext_ack *extack)
3104 {
3105 	struct cake_sched_data *q = qdisc_priv(sch);
3106 
3107 	if (cl)
3108 		return NULL;
3109 	return q->block;
3110 }
3111 
3112 static int cake_dump_class(struct Qdisc *sch, unsigned long cl,
3113 			   struct sk_buff *skb, struct tcmsg *tcm)
3114 {
3115 	tcm->tcm_handle |= TC_H_MIN(cl);
3116 	return 0;
3117 }
3118 
3119 static int cake_dump_class_stats(struct Qdisc *sch, unsigned long cl,
3120 				 struct gnet_dump *d)
3121 {
3122 	struct cake_sched_data *q = qdisc_priv(sch);
3123 	const struct cake_flow *flow = NULL;
3124 	struct gnet_stats_queue qs = { 0 };
3125 	struct nlattr *stats;
3126 	u32 idx = cl - 1;
3127 
3128 	if (idx < CAKE_QUEUES * q->tin_cnt) {
3129 		const struct cake_tin_data *b = \
3130 			&q->tins[q->tin_order[idx / CAKE_QUEUES]];
3131 		const struct sk_buff *skb;
3132 
3133 		flow = &b->flows[idx % CAKE_QUEUES];
3134 
3135 		if (flow->head) {
3136 			sch_tree_lock(sch);
3137 			skb = flow->head;
3138 			while (skb) {
3139 				qs.qlen++;
3140 				skb = skb->next;
3141 			}
3142 			sch_tree_unlock(sch);
3143 		}
3144 		qs.backlog = b->backlogs[idx % CAKE_QUEUES];
3145 		qs.drops = flow->dropped;
3146 	}
3147 	if (gnet_stats_copy_queue(d, NULL, &qs, qs.qlen) < 0)
3148 		return -1;
3149 	if (flow) {
3150 		ktime_t now = ktime_get();
3151 
3152 		stats = nla_nest_start_noflag(d->skb, TCA_STATS_APP);
3153 		if (!stats)
3154 			return -1;
3155 
3156 #define PUT_STAT_U32(attr, data) do {				       \
3157 		if (nla_put_u32(d->skb, TCA_CAKE_STATS_ ## attr, data)) \
3158 			goto nla_put_failure;			       \
3159 	} while (0)
3160 #define PUT_STAT_S32(attr, data) do {				       \
3161 		if (nla_put_s32(d->skb, TCA_CAKE_STATS_ ## attr, data)) \
3162 			goto nla_put_failure;			       \
3163 	} while (0)
3164 
3165 		PUT_STAT_S32(DEFICIT, flow->deficit);
3166 		PUT_STAT_U32(DROPPING, flow->cvars.dropping);
3167 		PUT_STAT_U32(COBALT_COUNT, flow->cvars.count);
3168 		PUT_STAT_U32(P_DROP, flow->cvars.p_drop);
3169 		if (flow->cvars.p_drop) {
3170 			PUT_STAT_S32(BLUE_TIMER_US,
3171 				     ktime_to_us(
3172 					     ktime_sub(now,
3173 						       flow->cvars.blue_timer)));
3174 		}
3175 		if (flow->cvars.dropping) {
3176 			PUT_STAT_S32(DROP_NEXT_US,
3177 				     ktime_to_us(
3178 					     ktime_sub(now,
3179 						       flow->cvars.drop_next)));
3180 		}
3181 
3182 		if (nla_nest_end(d->skb, stats) < 0)
3183 			return -1;
3184 	}
3185 
3186 	return 0;
3187 
3188 nla_put_failure:
3189 	nla_nest_cancel(d->skb, stats);
3190 	return -1;
3191 }
3192 
3193 static void cake_walk(struct Qdisc *sch, struct qdisc_walker *arg)
3194 {
3195 	struct cake_sched_data *q = qdisc_priv(sch);
3196 	unsigned int i, j;
3197 
3198 	if (arg->stop)
3199 		return;
3200 
3201 	for (i = 0; i < q->tin_cnt; i++) {
3202 		struct cake_tin_data *b = &q->tins[q->tin_order[i]];
3203 
3204 		for (j = 0; j < CAKE_QUEUES; j++) {
3205 			if (list_empty(&b->flows[j].flowchain)) {
3206 				arg->count++;
3207 				continue;
3208 			}
3209 			if (!tc_qdisc_stats_dump(sch, i * CAKE_QUEUES + j + 1,
3210 						 arg))
3211 				break;
3212 		}
3213 	}
3214 }
3215 
3216 static const struct Qdisc_class_ops cake_class_ops = {
3217 	.leaf		=	cake_leaf,
3218 	.find		=	cake_find,
3219 	.tcf_block	=	cake_tcf_block,
3220 	.bind_tcf	=	cake_bind,
3221 	.unbind_tcf	=	cake_unbind,
3222 	.dump		=	cake_dump_class,
3223 	.dump_stats	=	cake_dump_class_stats,
3224 	.walk		=	cake_walk,
3225 };
3226 
3227 static struct Qdisc_ops cake_qdisc_ops __read_mostly = {
3228 	.cl_ops		=	&cake_class_ops,
3229 	.id		=	"cake",
3230 	.priv_size	=	sizeof(struct cake_sched_data),
3231 	.enqueue	=	cake_enqueue,
3232 	.dequeue	=	cake_dequeue,
3233 	.peek		=	qdisc_peek_dequeued,
3234 	.init		=	cake_init,
3235 	.reset		=	cake_reset,
3236 	.destroy	=	cake_destroy,
3237 	.change		=	cake_change,
3238 	.dump		=	cake_dump,
3239 	.dump_stats	=	cake_dump_stats,
3240 	.owner		=	THIS_MODULE,
3241 };
3242 MODULE_ALIAS_NET_SCH("cake");
3243 
3244 struct cake_mq_sched {
3245 	struct mq_sched mq_priv; /* must be first */
3246 	struct cake_sched_config cake_config;
3247 };
3248 
3249 static void cake_mq_destroy(struct Qdisc *sch)
3250 {
3251 	mq_destroy_common(sch);
3252 }
3253 
3254 static int cake_mq_init(struct Qdisc *sch, struct nlattr *opt,
3255 			struct netlink_ext_ack *extack)
3256 {
3257 	struct cake_mq_sched *priv = qdisc_priv(sch);
3258 	struct net_device *dev = qdisc_dev(sch);
3259 	int ret, ntx;
3260 	bool _unused;
3261 
3262 	cake_config_init(&priv->cake_config, true);
3263 	if (opt) {
3264 		ret = cake_config_change(&priv->cake_config, opt, extack, &_unused);
3265 		if (ret)
3266 			return ret;
3267 	}
3268 
3269 	ret = mq_init_common(sch, opt, extack, &cake_qdisc_ops);
3270 	if (ret)
3271 		return ret;
3272 
3273 	for (ntx = 0; ntx < dev->num_tx_queues; ntx++)
3274 		cake_config_replace(priv->mq_priv.qdiscs[ntx], &priv->cake_config);
3275 
3276 	return 0;
3277 }
3278 
3279 static int cake_mq_dump(struct Qdisc *sch, struct sk_buff *skb)
3280 {
3281 	struct cake_mq_sched *priv = qdisc_priv(sch);
3282 
3283 	mq_dump_common(sch, skb);
3284 	return cake_config_dump(&priv->cake_config, skb);
3285 }
3286 
3287 static int cake_mq_change(struct Qdisc *sch, struct nlattr *opt,
3288 			  struct netlink_ext_ack *extack)
3289 {
3290 	struct cake_mq_sched *priv = qdisc_priv(sch);
3291 	struct net_device *dev = qdisc_dev(sch);
3292 	bool overhead_changed = false;
3293 	unsigned int ntx;
3294 	int ret;
3295 
3296 	ret = cake_config_change(&priv->cake_config, opt, extack, &overhead_changed);
3297 	if (ret)
3298 		return ret;
3299 
3300 	for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
3301 		struct Qdisc *chld = rtnl_dereference(netdev_get_tx_queue(dev, ntx)->qdisc_sleeping);
3302 		struct cake_sched_data *qd = qdisc_priv(chld);
3303 
3304 		if (overhead_changed) {
3305 			qd->max_netlen = 0;
3306 			qd->max_adjlen = 0;
3307 			qd->min_netlen = ~0;
3308 			qd->min_adjlen = ~0;
3309 		}
3310 
3311 		if (qd->tins) {
3312 			sch_tree_lock(chld);
3313 			cake_reconfigure(chld);
3314 			sch_tree_unlock(chld);
3315 		}
3316 	}
3317 
3318 	return 0;
3319 }
3320 
3321 static int cake_mq_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
3322 			 struct Qdisc **old, struct netlink_ext_ack *extack)
3323 {
3324 	NL_SET_ERR_MSG(extack, "can't replace cake_mq sub-qdiscs");
3325 	return -EOPNOTSUPP;
3326 }
3327 
3328 static const struct Qdisc_class_ops cake_mq_class_ops = {
3329 	.select_queue	= mq_select_queue,
3330 	.graft		= cake_mq_graft,
3331 	.leaf		= mq_leaf,
3332 	.find		= mq_find,
3333 	.walk		= mq_walk,
3334 	.dump		= mq_dump_class,
3335 	.dump_stats	= mq_dump_class_stats,
3336 };
3337 
3338 static struct Qdisc_ops cake_mq_qdisc_ops __read_mostly = {
3339 	.cl_ops		=	&cake_mq_class_ops,
3340 	.id		=	"cake_mq",
3341 	.priv_size	=	sizeof(struct cake_mq_sched),
3342 	.init		=	cake_mq_init,
3343 	.destroy	=	cake_mq_destroy,
3344 	.attach		=	mq_attach,
3345 	.change		=	cake_mq_change,
3346 	.change_real_num_tx = mq_change_real_num_tx,
3347 	.dump		=	cake_mq_dump,
3348 	.owner		=	THIS_MODULE,
3349 };
3350 MODULE_ALIAS_NET_SCH("cake_mq");
3351 
3352 static int __init cake_module_init(void)
3353 {
3354 	int ret;
3355 
3356 	ret = register_qdisc(&cake_qdisc_ops);
3357 	if (ret)
3358 		return ret;
3359 
3360 	ret = register_qdisc(&cake_mq_qdisc_ops);
3361 	if (ret)
3362 		unregister_qdisc(&cake_qdisc_ops);
3363 
3364 	return ret;
3365 }
3366 
3367 static void __exit cake_module_exit(void)
3368 {
3369 	unregister_qdisc(&cake_qdisc_ops);
3370 	unregister_qdisc(&cake_mq_qdisc_ops);
3371 }
3372 
3373 module_init(cake_module_init)
3374 module_exit(cake_module_exit)
3375 MODULE_AUTHOR("Jonathan Morton");
3376 MODULE_LICENSE("Dual BSD/GPL");
3377 MODULE_DESCRIPTION("The CAKE shaper.");
3378 MODULE_IMPORT_NS("NET_SCHED_INTERNAL");
3379