projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
lib/idr.c: use kmem_cache_zalloc() for the idr_layer cache
[linux-2.6-block.git] / net / sched / sch_hfsc.c
CommitLineData
1da177e4
LT		 1/*
2 * Copyright (c) 2003 Patrick McHardy, <kaber@trash.net>
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version 2
7 * of the License, or (at your option) any later version.
8 *
9 * 2003-10-17 - Ported from altq
10 */
11/*
12 * Copyright (c) 1997-1999 Carnegie Mellon University. All Rights Reserved.
13 *
14 * Permission to use, copy, modify, and distribute this software and
15 * its documentation is hereby granted (including for commercial or
16 * for-profit use), provided that both the copyright notice and this
17 * permission notice appear in all copies of the software, derivative
18 * works, or modified versions, and any portions thereof.
19 *
20 * THIS SOFTWARE IS EXPERIMENTAL AND IS KNOWN TO HAVE BUGS, SOME OF
21 * WHICH MAY HAVE SERIOUS CONSEQUENCES. CARNEGIE MELLON PROVIDES THIS
22 * SOFTWARE IN ITS ``AS IS'' CONDITION, AND ANY EXPRESS OR IMPLIED
23 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
25 * DISCLAIMED. IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
28 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
29 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
30 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
32 * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
33 * DAMAGE.
34 *
35 * Carnegie Mellon encourages (but does not require) users of this
36 * software to return any improvements or extensions that they make,
37 * and to grant Carnegie Mellon the rights to redistribute these
38 * changes without encumbrance.
39 */
40/*
41 * H-FSC is described in Proceedings of SIGCOMM'97,
42 * "A Hierarchical Fair Service Curve Algorithm for Link-Sharing,
43 * Real-Time and Priority Service"
44 * by Ion Stoica, Hui Zhang, and T. S. Eugene Ng.
45 *
46 * Oleg Cherevko <olwi@aq.ml.com.ua> added the upperlimit for link-sharing.
47 * when a class has an upperlimit, the fit-time is computed from the
48 * upperlimit service curve. the link-sharing scheduler does not schedule
49 * a class whose fit-time exceeds the current time.
50 */
51
52#include <linux/kernel.h>
1da177e4
LT		 53#include <linux/module.h>
54#include <linux/types.h>
55#include <linux/errno.h>
1da177e4
LT		 56#include <linux/compiler.h>
57#include <linux/spinlock.h>
58#include <linux/skbuff.h>
59#include <linux/string.h>
60#include <linux/slab.h>
1da177e4
LT		 61#include <linux/list.h>
62#include <linux/rbtree.h>
63#include <linux/init.h>
1da177e4
LT		 64#include <linux/rtnetlink.h>
65#include <linux/pkt_sched.h>
dc5fc579 66#include <net/netlink.h>
1da177e4
LT		 67#include <net/pkt_sched.h>
68#include <net/pkt_cls.h>
1da177e4
LT		 69#include <asm/div64.h>
70
1da177e4
LT		 71/*
72 * kernel internal service curve representation:
73 * coordinates are given by 64 bit unsigned integers.
74 * x-axis: unit is clock count.
75 * y-axis: unit is byte.
76 *
77 * The service curve parameters are converted to the internal
78 * representation. The slope values are scaled to avoid overflow.
79 * the inverse slope values as well as the y-projection of the 1st
80 * segment are kept in order to to avoid 64-bit divide operations
81 * that are expensive on 32-bit architectures.
82 */
83
84struct internal_sc
85{
86 u64 sm1; /* scaled slope of the 1st segment */
87 u64 ism1; /* scaled inverse-slope of the 1st segment */
88 u64 dx; /* the x-projection of the 1st segment */
89 u64 dy; /* the y-projection of the 1st segment */
90 u64 sm2; /* scaled slope of the 2nd segment */
91 u64 ism2; /* scaled inverse-slope of the 2nd segment */
92};
93
94/* runtime service curve */
95struct runtime_sc
96{
97 u64 x; /* current starting position on x-axis */
98 u64 y; /* current starting position on y-axis */
99 u64 sm1; /* scaled slope of the 1st segment */
100 u64 ism1; /* scaled inverse-slope of the 1st segment */
101 u64 dx; /* the x-projection of the 1st segment */
102 u64 dy; /* the y-projection of the 1st segment */
103 u64 sm2; /* scaled slope of the 2nd segment */
104 u64 ism2; /* scaled inverse-slope of the 2nd segment */
105};
106
107enum hfsc_class_flags
108{
109 HFSC_RSC = 0x1,
110 HFSC_FSC = 0x2,
111 HFSC_USC = 0x4
112};
113
114struct hfsc_class
115{
be0d39d5 116 struct Qdisc_class_common cl_common;
1da177e4
LT		 117 unsigned int refcnt; /* usage count */
118
119 struct gnet_stats_basic bstats;
120 struct gnet_stats_queue qstats;
121 struct gnet_stats_rate_est rate_est;
1da177e4
LT		 122 unsigned int level; /* class level in hierarchy */
123 struct tcf_proto *filter_list; /* filter list */
124 unsigned int filter_cnt; /* filter count */
125
126 struct hfsc_sched *sched; /* scheduler data */
127 struct hfsc_class *cl_parent; /* parent class */
128 struct list_head siblings; /* sibling classes */
129 struct list_head children; /* child classes */
130 struct Qdisc *qdisc; /* leaf qdisc */
131
132 struct rb_node el_node; /* qdisc's eligible tree member */
133 struct rb_root vt_tree; /* active children sorted by cl_vt */
134 struct rb_node vt_node; /* parent's vt_tree member */
135 struct rb_root cf_tree; /* active children sorted by cl_f */
136 struct rb_node cf_node; /* parent's cf_heap member */
1da177e4
LT		 137 struct list_head dlist; /* drop list member */
138
139 u64 cl_total; /* total work in bytes */
140 u64 cl_cumul; /* cumulative work in bytes done by
141 real-time criteria */
142
143 u64 cl_d; /* deadline*/
144 u64 cl_e; /* eligible time */
145 u64 cl_vt; /* virtual time */
146 u64 cl_f; /* time when this class will fit for
147 link-sharing, max(myf, cfmin) */
148 u64 cl_myf; /* my fit-time (calculated from this
149 class's own upperlimit curve) */
150 u64 cl_myfadj; /* my fit-time adjustment (to cancel
151 history dependence) */
152 u64 cl_cfmin; /* earliest children's fit-time (used
153 with cl_myf to obtain cl_f) */
154 u64 cl_cvtmin; /* minimal virtual time among the
155 children fit for link-sharing
156 (monotonic within a period) */
157 u64 cl_vtadj; /* intra-period cumulative vt
158 adjustment */
159 u64 cl_vtoff; /* inter-period cumulative vt offset */
160 u64 cl_cvtmax; /* max child's vt in the last period */
161 u64 cl_cvtoff; /* cumulative cvtmax of all periods */
9a94b351 162 u64 cl_pcvtoff; /* parent's cvtoff at initialization
1da177e4
LT		 163 time */
164
165 struct internal_sc cl_rsc; /* internal real-time service curve */
166 struct internal_sc cl_fsc; /* internal fair service curve */
167 struct internal_sc cl_usc; /* internal upperlimit service curve */
168 struct runtime_sc cl_deadline; /* deadline curve */
169 struct runtime_sc cl_eligible; /* eligible curve */
170 struct runtime_sc cl_virtual; /* virtual curve */
171 struct runtime_sc cl_ulimit; /* upperlimit curve */
172
173 unsigned long cl_flags; /* which curves are valid */
174 unsigned long cl_vtperiod; /* vt period sequence number */
175 unsigned long cl_parentperiod;/* parent's vt period sequence number*/
176 unsigned long cl_nactive; /* number of active children */
177};
178
1da177e4
LT		 179struct hfsc_sched
180{
181 u16 defcls; /* default class id */
182 struct hfsc_class root; /* root class */
be0d39d5 183 struct Qdisc_class_hash clhash; /* class hash */
1da177e4
LT		 184 struct rb_root eligible; /* eligible tree */
185 struct list_head droplist; /* active leaf class list (for
186 dropping) */
ed2b229a 187 struct qdisc_watchdog watchdog; /* watchdog timer */
1da177e4
LT		 188};
189
1da177e4
LT		 190#define HT_INFINITY 0xffffffffffffffffULL /* infinite time value */
191
192
193/*
194 * eligible tree holds backlogged classes being sorted by their eligible times.
195 * there is one eligible tree per hfsc instance.
196 */
197
198static void
199eltree_insert(struct hfsc_class *cl)
200{
201 struct rb_node **p = &cl->sched->eligible.rb_node;
202 struct rb_node *parent = NULL;
203 struct hfsc_class *cl1;
204
205 while (*p != NULL) {
206 parent = *p;
207 cl1 = rb_entry(parent, struct hfsc_class, el_node);
208 if (cl->cl_e >= cl1->cl_e)
209 p = &parent->rb_right;
210 else
211 p = &parent->rb_left;
212 }
213 rb_link_node(&cl->el_node, parent, p);
214 rb_insert_color(&cl->el_node, &cl->sched->eligible);
215}
216
217static inline void
218eltree_remove(struct hfsc_class *cl)
219{
220 rb_erase(&cl->el_node, &cl->sched->eligible);
221}
222
223static inline void
224eltree_update(struct hfsc_class *cl)
225{
226 eltree_remove(cl);
227 eltree_insert(cl);
228}
229
230/* find the class with the minimum deadline among the eligible classes */
231static inline struct hfsc_class *
232eltree_get_mindl(struct hfsc_sched *q, u64 cur_time)
233{
234 struct hfsc_class *p, *cl = NULL;
235 struct rb_node *n;
236
237 for (n = rb_first(&q->eligible); n != NULL; n = rb_next(n)) {
238 p = rb_entry(n, struct hfsc_class, el_node);
239 if (p->cl_e > cur_time)
240 break;
241 if (cl == NULL || p->cl_d < cl->cl_d)
242 cl = p;
243 }
244 return cl;
245}
246
247/* find the class with minimum eligible time among the eligible classes */
248static inline struct hfsc_class *
249eltree_get_minel(struct hfsc_sched *q)
250{
251 struct rb_node *n;
10297b99 252
1da177e4
LT		 253 n = rb_first(&q->eligible);
254 if (n == NULL)
255 return NULL;
256 return rb_entry(n, struct hfsc_class, el_node);
257}
258
259/*
260 * vttree holds holds backlogged child classes being sorted by their virtual
261 * time. each intermediate class has one vttree.
262 */
263static void
264vttree_insert(struct hfsc_class *cl)
265{
266 struct rb_node **p = &cl->cl_parent->vt_tree.rb_node;
267 struct rb_node *parent = NULL;
268 struct hfsc_class *cl1;
269
270 while (*p != NULL) {
271 parent = *p;
272 cl1 = rb_entry(parent, struct hfsc_class, vt_node);
273 if (cl->cl_vt >= cl1->cl_vt)
274 p = &parent->rb_right;
275 else
276 p = &parent->rb_left;
277 }
278 rb_link_node(&cl->vt_node, parent, p);
279 rb_insert_color(&cl->vt_node, &cl->cl_parent->vt_tree);
280}
281
282static inline void
283vttree_remove(struct hfsc_class *cl)
284{
285 rb_erase(&cl->vt_node, &cl->cl_parent->vt_tree);
286}
287
288static inline void
289vttree_update(struct hfsc_class *cl)
290{
291 vttree_remove(cl);
292 vttree_insert(cl);
293}
294
295static inline struct hfsc_class *
296vttree_firstfit(struct hfsc_class *cl, u64 cur_time)
297{
298 struct hfsc_class *p;
299 struct rb_node *n;
300
301 for (n = rb_first(&cl->vt_tree); n != NULL; n = rb_next(n)) {
302 p = rb_entry(n, struct hfsc_class, vt_node);
303 if (p->cl_f <= cur_time)
304 return p;
305 }
306 return NULL;
307}
308
309/*
310 * get the leaf class with the minimum vt in the hierarchy
311 */
312static struct hfsc_class *
313vttree_get_minvt(struct hfsc_class *cl, u64 cur_time)
314{
315 /* if root-class's cfmin is bigger than cur_time nothing to do */
316 if (cl->cl_cfmin > cur_time)
317 return NULL;
318
319 while (cl->level > 0) {
320 cl = vttree_firstfit(cl, cur_time);
321 if (cl == NULL)
322 return NULL;
323 /*
324 * update parent's cl_cvtmin.
325 */
326 if (cl->cl_parent->cl_cvtmin < cl->cl_vt)
327 cl->cl_parent->cl_cvtmin = cl->cl_vt;
328 }
329 return cl;
330}
331
332static void
333cftree_insert(struct hfsc_class *cl)
334{
335 struct rb_node **p = &cl->cl_parent->cf_tree.rb_node;
336 struct rb_node *parent = NULL;
337 struct hfsc_class *cl1;
338
339 while (*p != NULL) {
340 parent = *p;
341 cl1 = rb_entry(parent, struct hfsc_class, cf_node);
342 if (cl->cl_f >= cl1->cl_f)
343 p = &parent->rb_right;
344 else
345 p = &parent->rb_left;
346 }
347 rb_link_node(&cl->cf_node, parent, p);
348 rb_insert_color(&cl->cf_node, &cl->cl_parent->cf_tree);
349}
350
351static inline void
352cftree_remove(struct hfsc_class *cl)
353{
354 rb_erase(&cl->cf_node, &cl->cl_parent->cf_tree);
355}
356
357static inline void
358cftree_update(struct hfsc_class *cl)
359{
360 cftree_remove(cl);
361 cftree_insert(cl);
362}
363
364/*
365 * service curve support functions
366 *
367 * external service curve parameters
368 * m: bps
369 * d: us
370 * internal service curve parameters
371 * sm: (bytes/psched_us) << SM_SHIFT
372 * ism: (psched_us/byte) << ISM_SHIFT
373 * dx: psched_us
374 *
641b9e0e 375 * The clock source resolution with ktime is 1.024us.
1da177e4
LT		 376 *
377 * sm and ism are scaled in order to keep effective digits.
378 * SM_SHIFT and ISM_SHIFT are selected to keep at least 4 effective
379 * digits in decimal using the following table.
380 *
1da177e4
LT		 381 * bits/sec 100Kbps 1Mbps 10Mbps 100Mbps 1Gbps
382 * ------------+-------------------------------------------------------
641b9e0e 383 * bytes/1.024us 12.8e-3 128e-3 1280e-3 12800e-3 128000e-3
1da177e4 384 *
641b9e0e 385 * 1.024us/byte 78.125 7.8125 0.78125 0.078125 0.0078125
1da177e4
LT		 386 */
387#define SM_SHIFT 20
388#define ISM_SHIFT 18
389
390#define SM_MASK ((1ULL << SM_SHIFT) - 1)
391#define ISM_MASK ((1ULL << ISM_SHIFT) - 1)
392
393static inline u64
394seg_x2y(u64 x, u64 sm)
395{
396 u64 y;
397
398 /*
399 * compute
400 * y = x * sm >> SM_SHIFT
401 * but divide it for the upper and lower bits to avoid overflow
402 */
403 y = (x >> SM_SHIFT) * sm + (((x & SM_MASK) * sm) >> SM_SHIFT);
404 return y;
405}
406
407static inline u64
408seg_y2x(u64 y, u64 ism)
409{
410 u64 x;
411
412 if (y == 0)
413 x = 0;
414 else if (ism == HT_INFINITY)
415 x = HT_INFINITY;
416 else {
417 x = (y >> ISM_SHIFT) * ism
418 + (((y & ISM_MASK) * ism) >> ISM_SHIFT);
419 }
420 return x;
421}
422
423/* Convert m (bps) into sm (bytes/psched us) */
424static u64
425m2sm(u32 m)
426{
427 u64 sm;
428
429 sm = ((u64)m << SM_SHIFT);
00c04af9
PM		 430 sm += PSCHED_TICKS_PER_SEC - 1;
431 do_div(sm, PSCHED_TICKS_PER_SEC);
1da177e4
LT		 432 return sm;
433}
434
435/* convert m (bps) into ism (psched us/byte) */
436static u64
437m2ism(u32 m)
438{
439 u64 ism;
440
441 if (m == 0)
442 ism = HT_INFINITY;
443 else {
00c04af9 444 ism = ((u64)PSCHED_TICKS_PER_SEC << ISM_SHIFT);
1da177e4
LT		 445 ism += m - 1;
446 do_div(ism, m);
447 }
448 return ism;
449}
450
451/* convert d (us) into dx (psched us) */
452static u64
453d2dx(u32 d)
454{
455 u64 dx;
456
00c04af9 457 dx = ((u64)d * PSCHED_TICKS_PER_SEC);
538e43a4
PM		 458 dx += USEC_PER_SEC - 1;
459 do_div(dx, USEC_PER_SEC);
1da177e4
LT		 460 return dx;
461}
462
463/* convert sm (bytes/psched us) into m (bps) */
464static u32
465sm2m(u64 sm)
466{
467 u64 m;
468
00c04af9 469 m = (sm * PSCHED_TICKS_PER_SEC) >> SM_SHIFT;
1da177e4
LT		 470 return (u32)m;
471}
472
473/* convert dx (psched us) into d (us) */
474static u32
475dx2d(u64 dx)
476{
477 u64 d;
478
538e43a4 479 d = dx * USEC_PER_SEC;
00c04af9 480 do_div(d, PSCHED_TICKS_PER_SEC);
1da177e4
LT		 481 return (u32)d;
482}
483
484static void
485sc2isc(struct tc_service_curve *sc, struct internal_sc *isc)
486{
487 isc->sm1 = m2sm(sc->m1);
488 isc->ism1 = m2ism(sc->m1);
489 isc->dx = d2dx(sc->d);
490 isc->dy = seg_x2y(isc->dx, isc->sm1);
491 isc->sm2 = m2sm(sc->m2);
492 isc->ism2 = m2ism(sc->m2);
493}
494
495/*
496 * initialize the runtime service curve with the given internal
497 * service curve starting at (x, y).
498 */
499static void
500rtsc_init(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
501{
502 rtsc->x = x;
503 rtsc->y = y;
504 rtsc->sm1 = isc->sm1;
505 rtsc->ism1 = isc->ism1;
506 rtsc->dx = isc->dx;
507 rtsc->dy = isc->dy;
508 rtsc->sm2 = isc->sm2;
509 rtsc->ism2 = isc->ism2;
510}
511
512/*
513 * calculate the y-projection of the runtime service curve by the
514 * given x-projection value
515 */
516static u64
517rtsc_y2x(struct runtime_sc *rtsc, u64 y)
518{
519 u64 x;
520
521 if (y < rtsc->y)
522 x = rtsc->x;
523 else if (y <= rtsc->y + rtsc->dy) {
524 /* x belongs to the 1st segment */
525 if (rtsc->dy == 0)
526 x = rtsc->x + rtsc->dx;
527 else
528 x = rtsc->x + seg_y2x(y - rtsc->y, rtsc->ism1);
529 } else {
530 /* x belongs to the 2nd segment */
531 x = rtsc->x + rtsc->dx
532 + seg_y2x(y - rtsc->y - rtsc->dy, rtsc->ism2);
533 }
534 return x;
535}
536
537static u64
538rtsc_x2y(struct runtime_sc *rtsc, u64 x)
539{
540 u64 y;
541
542 if (x <= rtsc->x)
543 y = rtsc->y;
544 else if (x <= rtsc->x + rtsc->dx)
545 /* y belongs to the 1st segment */
546 y = rtsc->y + seg_x2y(x - rtsc->x, rtsc->sm1);
547 else
548 /* y belongs to the 2nd segment */
549 y = rtsc->y + rtsc->dy
550 + seg_x2y(x - rtsc->x - rtsc->dx, rtsc->sm2);
551 return y;
552}
553
554/*
555 * update the runtime service curve by taking the minimum of the current
556 * runtime service curve and the service curve starting at (x, y).
557 */
558static void
559rtsc_min(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
560{
561 u64 y1, y2, dx, dy;
562 u32 dsm;
563
564 if (isc->sm1 <= isc->sm2) {
565 /* service curve is convex */
566 y1 = rtsc_x2y(rtsc, x);
567 if (y1 < y)
568 /* the current rtsc is smaller */
569 return;
570 rtsc->x = x;
571 rtsc->y = y;
572 return;
573 }
574
575 /*
576 * service curve is concave
577 * compute the two y values of the current rtsc
578 * y1: at x
579 * y2: at (x + dx)
580 */
581 y1 = rtsc_x2y(rtsc, x);
582 if (y1 <= y) {
583 /* rtsc is below isc, no change to rtsc */
584 return;
585 }
586
587 y2 = rtsc_x2y(rtsc, x + isc->dx);
588 if (y2 >= y + isc->dy) {
589 /* rtsc is above isc, replace rtsc by isc */
590 rtsc->x = x;
591 rtsc->y = y;
592 rtsc->dx = isc->dx;
593 rtsc->dy = isc->dy;
594 return;
595 }
596
597 /*
598 * the two curves intersect
599 * compute the offsets (dx, dy) using the reverse
600 * function of seg_x2y()
601 * seg_x2y(dx, sm1) == seg_x2y(dx, sm2) + (y1 - y)
602 */
603 dx = (y1 - y) << SM_SHIFT;
604 dsm = isc->sm1 - isc->sm2;
605 do_div(dx, dsm);
606 /*
607 * check if (x, y1) belongs to the 1st segment of rtsc.
608 * if so, add the offset.
609 */
610 if (rtsc->x + rtsc->dx > x)
611 dx += rtsc->x + rtsc->dx - x;
612 dy = seg_x2y(dx, isc->sm1);
613
614 rtsc->x = x;
615 rtsc->y = y;
616 rtsc->dx = dx;
617 rtsc->dy = dy;
618 return;
619}
620
621static void
622init_ed(struct hfsc_class *cl, unsigned int next_len)
623{
3bebcda2 624 u64 cur_time = psched_get_time();
1da177e4
LT		 625
626 /* update the deadline curve */
627 rtsc_min(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);
628
629 /*
630 * update the eligible curve.
631 * for concave, it is equal to the deadline curve.
632 * for convex, it is a linear curve with slope m2.
633 */
634 cl->cl_eligible = cl->cl_deadline;
635 if (cl->cl_rsc.sm1 <= cl->cl_rsc.sm2) {
636 cl->cl_eligible.dx = 0;
637 cl->cl_eligible.dy = 0;
638 }
639
640 /* compute e and d */
641 cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
642 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
643
644 eltree_insert(cl);
645}
646
647static void
648update_ed(struct hfsc_class *cl, unsigned int next_len)
649{
650 cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
651 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
652
653 eltree_update(cl);
654}
655
656static inline void
657update_d(struct hfsc_class *cl, unsigned int next_len)
658{
659 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
660}
661
662static inline void
663update_cfmin(struct hfsc_class *cl)
664{
665 struct rb_node *n = rb_first(&cl->cf_tree);
666 struct hfsc_class *p;
667
668 if (n == NULL) {
669 cl->cl_cfmin = 0;
670 return;
671 }
672 p = rb_entry(n, struct hfsc_class, cf_node);
673 cl->cl_cfmin = p->cl_f;
674}
675
676static void
677init_vf(struct hfsc_class *cl, unsigned int len)
678{
679 struct hfsc_class *max_cl;
680 struct rb_node *n;
681 u64 vt, f, cur_time;
682 int go_active;
683
684 cur_time = 0;
685 go_active = 1;
686 for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
687 if (go_active && cl->cl_nactive++ == 0)
688 go_active = 1;
689 else
690 go_active = 0;
691
692 if (go_active) {
693 n = rb_last(&cl->cl_parent->vt_tree);
694 if (n != NULL) {
695 max_cl = rb_entry(n, struct hfsc_class,vt_node);
696 /*
697 * set vt to the average of the min and max
698 * classes. if the parent's period didn't
699 * change, don't decrease vt of the class.
700 */
701 vt = max_cl->cl_vt;
702 if (cl->cl_parent->cl_cvtmin != 0)
703 vt = (cl->cl_parent->cl_cvtmin + vt)/2;
704
705 if (cl->cl_parent->cl_vtperiod !=
706 cl->cl_parentperiod || vt > cl->cl_vt)
707 cl->cl_vt = vt;
708 } else {
709 /*
710 * first child for a new parent backlog period.
711 * add parent's cvtmax to cvtoff to make a new
712 * vt (vtoff + vt) larger than the vt in the
713 * last period for all children.
714 */
715 vt = cl->cl_parent->cl_cvtmax;
716 cl->cl_parent->cl_cvtoff += vt;
717 cl->cl_parent->cl_cvtmax = 0;
718 cl->cl_parent->cl_cvtmin = 0;
719 cl->cl_vt = 0;
720 }
721
722 cl->cl_vtoff = cl->cl_parent->cl_cvtoff -
723 cl->cl_pcvtoff;
724
725 /* update the virtual curve */
726 vt = cl->cl_vt + cl->cl_vtoff;
727 rtsc_min(&cl->cl_virtual, &cl->cl_fsc, vt,
10297b99 728 cl->cl_total);
1da177e4
LT		 729 if (cl->cl_virtual.x == vt) {
730 cl->cl_virtual.x -= cl->cl_vtoff;
731 cl->cl_vtoff = 0;
732 }
733 cl->cl_vtadj = 0;
734
735 cl->cl_vtperiod++; /* increment vt period */
736 cl->cl_parentperiod = cl->cl_parent->cl_vtperiod;
737 if (cl->cl_parent->cl_nactive == 0)
738 cl->cl_parentperiod++;
739 cl->cl_f = 0;
740
741 vttree_insert(cl);
742 cftree_insert(cl);
743
744 if (cl->cl_flags & HFSC_USC) {
745 /* class has upper limit curve */
746 if (cur_time == 0)
3bebcda2 747 cur_time = psched_get_time();
1da177e4
LT		 748
749 /* update the ulimit curve */
750 rtsc_min(&cl->cl_ulimit, &cl->cl_usc, cur_time,
10297b99 751 cl->cl_total);
1da177e4
LT		 752 /* compute myf */
753 cl->cl_myf = rtsc_y2x(&cl->cl_ulimit,
10297b99 754 cl->cl_total);
1da177e4
LT		 755 cl->cl_myfadj = 0;
756 }
757 }
758
759 f = max(cl->cl_myf, cl->cl_cfmin);
760 if (f != cl->cl_f) {
761 cl->cl_f = f;
762 cftree_update(cl);
763 update_cfmin(cl->cl_parent);
764 }
765 }
766}
767
768static void
769update_vf(struct hfsc_class *cl, unsigned int len, u64 cur_time)
770{
771 u64 f; /* , myf_bound, delta; */
772 int go_passive = 0;
773
774 if (cl->qdisc->q.qlen == 0 && cl->cl_flags & HFSC_FSC)
775 go_passive = 1;
776
777 for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
778 cl->cl_total += len;
779
780 if (!(cl->cl_flags & HFSC_FSC) || cl->cl_nactive == 0)
781 continue;
782
783 if (go_passive && --cl->cl_nactive == 0)
784 go_passive = 1;
785 else
786 go_passive = 0;
787
788 if (go_passive) {
789 /* no more active child, going passive */
790
791 /* update cvtmax of the parent class */
792 if (cl->cl_vt > cl->cl_parent->cl_cvtmax)
793 cl->cl_parent->cl_cvtmax = cl->cl_vt;
794
795 /* remove this class from the vt tree */
796 vttree_remove(cl);
797
798 cftree_remove(cl);
799 update_cfmin(cl->cl_parent);
800
801 continue;
802 }
803
804 /*
805 * update vt and f
806 */
807 cl->cl_vt = rtsc_y2x(&cl->cl_virtual, cl->cl_total)
10297b99 808 - cl->cl_vtoff + cl->cl_vtadj;
1da177e4
LT		 809
810 /*
811 * if vt of the class is smaller than cvtmin,
812 * the class was skipped in the past due to non-fit.
813 * if so, we need to adjust vtadj.
814 */
815 if (cl->cl_vt < cl->cl_parent->cl_cvtmin) {
816 cl->cl_vtadj += cl->cl_parent->cl_cvtmin - cl->cl_vt;
817 cl->cl_vt = cl->cl_parent->cl_cvtmin;
818 }
819
820 /* update the vt tree */
821 vttree_update(cl);
822
823 if (cl->cl_flags & HFSC_USC) {
824 cl->cl_myf = cl->cl_myfadj + rtsc_y2x(&cl->cl_ulimit,
10297b99 825 cl->cl_total);
1da177e4
LT		 826#if 0
827 /*
828 * This code causes classes to stay way under their
829 * limit when multiple classes are used at gigabit
830 * speed. needs investigation. -kaber
831 */
832 /*
833 * if myf lags behind by more than one clock tick
834 * from the current time, adjust myfadj to prevent
835 * a rate-limited class from going greedy.
836 * in a steady state under rate-limiting, myf
837 * fluctuates within one clock tick.
838 */
839 myf_bound = cur_time - PSCHED_JIFFIE2US(1);
840 if (cl->cl_myf < myf_bound) {
841 delta = cur_time - cl->cl_myf;
842 cl->cl_myfadj += delta;
843 cl->cl_myf += delta;
844 }
845#endif
846 }
847
848 f = max(cl->cl_myf, cl->cl_cfmin);
849 if (f != cl->cl_f) {
850 cl->cl_f = f;
851 cftree_update(cl);
852 update_cfmin(cl->cl_parent);
853 }
854 }
855}
856
857static void
858set_active(struct hfsc_class *cl, unsigned int len)
859{
860 if (cl->cl_flags & HFSC_RSC)
861 init_ed(cl, len);
862 if (cl->cl_flags & HFSC_FSC)
863 init_vf(cl, len);
864
865 list_add_tail(&cl->dlist, &cl->sched->droplist);
866}
867
868static void
869set_passive(struct hfsc_class *cl)
870{
871 if (cl->cl_flags & HFSC_RSC)
872 eltree_remove(cl);
873
874 list_del(&cl->dlist);
875
876 /*
877 * vttree is now handled in update_vf() so that update_vf(cl, 0, 0)
878 * needs to be called explicitly to remove a class from vttree.
879 */
880}
881
1da177e4
LT		 882static unsigned int
883qdisc_peek_len(struct Qdisc *sch)
884{
885 struct sk_buff *skb;
886 unsigned int len;
887
03c05f0d 888 skb = sch->ops->peek(sch);
1da177e4
LT		 889 if (skb == NULL) {
890 if (net_ratelimit())
891 printk("qdisc_peek_len: non work-conserving qdisc ?\n");
892 return 0;
893 }
0abf77e5 894 len = qdisc_pkt_len(skb);
03c05f0d 895
1da177e4
LT		 896 return len;
897}
898
899static void
900hfsc_purge_queue(struct Qdisc *sch, struct hfsc_class *cl)
901{
902 unsigned int len = cl->qdisc->q.qlen;
903
904 qdisc_reset(cl->qdisc);
f973b913 905 qdisc_tree_decrease_qlen(cl->qdisc, len);
1da177e4
LT		 906}
907
908static void
909hfsc_adjust_levels(struct hfsc_class *cl)
910{
911 struct hfsc_class *p;
912 unsigned int level;
913
914 do {
915 level = 0;
916 list_for_each_entry(p, &cl->children, siblings) {
210525d6
PM		 917 if (p->level >= level)
918 level = p->level + 1;
1da177e4 919 }
210525d6 920 cl->level = level;
1da177e4
LT		 921 } while ((cl = cl->cl_parent) != NULL);
922}
923
1da177e4
LT		 924static inline struct hfsc_class *
925hfsc_find_class(u32 classid, struct Qdisc *sch)
926{
927 struct hfsc_sched *q = qdisc_priv(sch);
be0d39d5 928 struct Qdisc_class_common *clc;
1da177e4 929
be0d39d5
PM		 930 clc = qdisc_class_find(&q->clhash, classid);
931 if (clc == NULL)
932 return NULL;
933 return container_of(clc, struct hfsc_class, cl_common);
1da177e4
LT		 934}
935
936static void
937hfsc_change_rsc(struct hfsc_class *cl, struct tc_service_curve *rsc,
10297b99 938 u64 cur_time)
1da177e4
LT		 939{
940 sc2isc(rsc, &cl->cl_rsc);
941 rtsc_init(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);
942 cl->cl_eligible = cl->cl_deadline;
943 if (cl->cl_rsc.sm1 <= cl->cl_rsc.sm2) {
944 cl->cl_eligible.dx = 0;
945 cl->cl_eligible.dy = 0;
946 }
947 cl->cl_flags |= HFSC_RSC;
948}
949
950static void
951hfsc_change_fsc(struct hfsc_class *cl, struct tc_service_curve *fsc)
952{
953 sc2isc(fsc, &cl->cl_fsc);
954 rtsc_init(&cl->cl_virtual, &cl->cl_fsc, cl->cl_vt, cl->cl_total);
955 cl->cl_flags |= HFSC_FSC;
956}
957
958static void
959hfsc_change_usc(struct hfsc_class *cl, struct tc_service_curve *usc,
10297b99 960 u64 cur_time)
1da177e4
LT		 961{
962 sc2isc(usc, &cl->cl_usc);
963 rtsc_init(&cl->cl_ulimit, &cl->cl_usc, cur_time, cl->cl_total);
964 cl->cl_flags |= HFSC_USC;
965}
966
27a3421e
PM		 967static const struct nla_policy hfsc_policy[TCA_HFSC_MAX + 1] = {
968 [TCA_HFSC_RSC] = { .len = sizeof(struct tc_service_curve) },
969 [TCA_HFSC_FSC] = { .len = sizeof(struct tc_service_curve) },
970 [TCA_HFSC_USC] = { .len = sizeof(struct tc_service_curve) },
971};
972
1da177e4
LT		 973static int
974hfsc_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
1e90474c 975 struct nlattr **tca, unsigned long *arg)
1da177e4
LT		 976{
977 struct hfsc_sched *q = qdisc_priv(sch);
978 struct hfsc_class *cl = (struct hfsc_class *)*arg;
979 struct hfsc_class *parent = NULL;
1e90474c
PM		 980 struct nlattr *opt = tca[TCA_OPTIONS];
981 struct nlattr *tb[TCA_HFSC_MAX + 1];
1da177e4
LT		 982 struct tc_service_curve *rsc = NULL, *fsc = NULL, *usc = NULL;
983 u64 cur_time;
cee63723 984 int err;
1da177e4 985
cee63723 986 if (opt == NULL)
1da177e4
LT		 987 return -EINVAL;
988
27a3421e 989 err = nla_parse_nested(tb, TCA_HFSC_MAX, opt, hfsc_policy);
cee63723
PM		 990 if (err < 0)
991 return err;
992
1e90474c 993 if (tb[TCA_HFSC_RSC]) {
1e90474c 994 rsc = nla_data(tb[TCA_HFSC_RSC]);
1da177e4
LT		 995 if (rsc->m1 == 0 && rsc->m2 == 0)
996 rsc = NULL;
997 }
998
1e90474c 999 if (tb[TCA_HFSC_FSC]) {
1e90474c 1000 fsc = nla_data(tb[TCA_HFSC_FSC]);
1da177e4
LT		 1001 if (fsc->m1 == 0 && fsc->m2 == 0)
1002 fsc = NULL;
1003 }
1004
1e90474c 1005 if (tb[TCA_HFSC_USC]) {
1e90474c 1006 usc = nla_data(tb[TCA_HFSC_USC]);
1da177e4
LT		 1007 if (usc->m1 == 0 && usc->m2 == 0)
1008 usc = NULL;
1009 }
1010
1011 if (cl != NULL) {
1012 if (parentid) {
be0d39d5
PM		 1013 if (cl->cl_parent &&
1014 cl->cl_parent->cl_common.classid != parentid)
1da177e4
LT		 1015 return -EINVAL;
1016 if (cl->cl_parent == NULL && parentid != TC_H_ROOT)
1017 return -EINVAL;
1018 }
3bebcda2 1019 cur_time = psched_get_time();
1da177e4 1020
71bcb09a
SH		 1021 if (tca[TCA_RATE]) {
1022 err = gen_replace_estimator(&cl->bstats, &cl->rate_est,
1023 qdisc_root_sleeping_lock(sch),
1024 tca[TCA_RATE]);
1025 if (err)
1026 return err;
1027 }
1028
1da177e4
LT		 1029 sch_tree_lock(sch);
1030 if (rsc != NULL)
1031 hfsc_change_rsc(cl, rsc, cur_time);
1032 if (fsc != NULL)
1033 hfsc_change_fsc(cl, fsc);
1034 if (usc != NULL)
1035 hfsc_change_usc(cl, usc, cur_time);
1036
1037 if (cl->qdisc->q.qlen != 0) {
1038 if (cl->cl_flags & HFSC_RSC)
1039 update_ed(cl, qdisc_peek_len(cl->qdisc));
1040 if (cl->cl_flags & HFSC_FSC)
1041 update_vf(cl, 0, cur_time);
1042 }
1043 sch_tree_unlock(sch);
1044
1da177e4
LT		 1045 return 0;
1046 }
1047
1048 if (parentid == TC_H_ROOT)
1049 return -EEXIST;
1050
1051 parent = &q->root;
1052 if (parentid) {
1053 parent = hfsc_find_class(parentid, sch);
1054 if (parent == NULL)
1055 return -ENOENT;
1056 }
1057
1058 if (classid == 0 || TC_H_MAJ(classid ^ sch->handle) != 0)
1059 return -EINVAL;
1060 if (hfsc_find_class(classid, sch))
1061 return -EEXIST;
1062
1063 if (rsc == NULL && fsc == NULL)
1064 return -EINVAL;
1065
0da974f4 1066 cl = kzalloc(sizeof(struct hfsc_class), GFP_KERNEL);
1da177e4
LT		 1067 if (cl == NULL)
1068 return -ENOBUFS;
1da177e4 1069
71bcb09a
SH		 1070 if (tca[TCA_RATE]) {
1071 err = gen_new_estimator(&cl->bstats, &cl->rate_est,
1072 qdisc_root_sleeping_lock(sch),
1073 tca[TCA_RATE]);
1074 if (err) {
1075 kfree(cl);
1076 return err;
1077 }
1078 }
1079
1da177e4
LT		 1080 if (rsc != NULL)
1081 hfsc_change_rsc(cl, rsc, 0);
1082 if (fsc != NULL)
1083 hfsc_change_fsc(cl, fsc);
1084 if (usc != NULL)
1085 hfsc_change_usc(cl, usc, 0);
1086
be0d39d5 1087 cl->cl_common.classid = classid;
1da177e4 1088 cl->refcnt = 1;
1da177e4
LT		 1089 cl->sched = q;
1090 cl->cl_parent = parent;
5ce2d488 1091 cl->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
bb949fbd 1092 &pfifo_qdisc_ops, classid);
1da177e4
LT		 1093 if (cl->qdisc == NULL)
1094 cl->qdisc = &noop_qdisc;
1da177e4
LT		 1095 INIT_LIST_HEAD(&cl->children);
1096 cl->vt_tree = RB_ROOT;
1097 cl->cf_tree = RB_ROOT;
1098
1099 sch_tree_lock(sch);
be0d39d5 1100 qdisc_class_hash_insert(&q->clhash, &cl->cl_common);
1da177e4
LT		 1101 list_add_tail(&cl->siblings, &parent->children);
1102 if (parent->level == 0)
1103 hfsc_purge_queue(sch, parent);
1104 hfsc_adjust_levels(parent);
1105 cl->cl_pcvtoff = parent->cl_cvtoff;
1106 sch_tree_unlock(sch);
1107
be0d39d5
PM		 1108 qdisc_class_hash_grow(sch, &q->clhash);
1109
1da177e4
LT		 1110 *arg = (unsigned long)cl;
1111 return 0;
1112}
1113
1da177e4
LT		 1114static void
1115hfsc_destroy_class(struct Qdisc *sch, struct hfsc_class *cl)
1116{
1117 struct hfsc_sched *q = qdisc_priv(sch);
1118
ff31ab56 1119 tcf_destroy_chain(&cl->filter_list);
1da177e4 1120 qdisc_destroy(cl->qdisc);
1da177e4 1121 gen_kill_estimator(&cl->bstats, &cl->rate_est);
1da177e4
LT		 1122 if (cl != &q->root)
1123 kfree(cl);
1124}
1125
1126static int
1127hfsc_delete_class(struct Qdisc *sch, unsigned long arg)
1128{
1129 struct hfsc_sched *q = qdisc_priv(sch);
1130 struct hfsc_class *cl = (struct hfsc_class *)arg;
1131
1132 if (cl->level > 0 || cl->filter_cnt > 0 || cl == &q->root)
1133 return -EBUSY;
1134
1135 sch_tree_lock(sch);
1136
1da177e4
LT		 1137 list_del(&cl->siblings);
1138 hfsc_adjust_levels(cl->cl_parent);
c38c83cb 1139
1da177e4 1140 hfsc_purge_queue(sch, cl);
be0d39d5 1141 qdisc_class_hash_remove(&q->clhash, &cl->cl_common);
c38c83cb 1142
1da177e4
LT		 1143 if (--cl->refcnt == 0)
1144 hfsc_destroy_class(sch, cl);
1145
1146 sch_tree_unlock(sch);
1147 return 0;
1148}
1149
1150static struct hfsc_class *
1151hfsc_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
1152{
1153 struct hfsc_sched *q = qdisc_priv(sch);
1154 struct hfsc_class *cl;
1155 struct tcf_result res;
1156 struct tcf_proto *tcf;
1157 int result;
1158
1159 if (TC_H_MAJ(skb->priority ^ sch->handle) == 0 &&
1160 (cl = hfsc_find_class(skb->priority, sch)) != NULL)
1161 if (cl->level == 0)
1162 return cl;
1163
c27f339a 1164 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
1da177e4
LT		 1165 tcf = q->root.filter_list;
1166 while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
1167#ifdef CONFIG_NET_CLS_ACT
1168 switch (result) {
1169 case TC_ACT_QUEUED:
10297b99 1170 case TC_ACT_STOLEN:
378a2f09 1171 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
10297b99 1172 case TC_ACT_SHOT:
1da177e4
LT		 1173 return NULL;
1174 }
1da177e4
LT		 1175#endif
1176 if ((cl = (struct hfsc_class *)res.class) == NULL) {
1177 if ((cl = hfsc_find_class(res.classid, sch)) == NULL)
1178 break; /* filter selected invalid classid */
1179 }
1180
1181 if (cl->level == 0)
1182 return cl; /* hit leaf class */
1183
1184 /* apply inner filter chain */
1185 tcf = cl->filter_list;
1186 }
1187
1188 /* classification failed, try default class */
1189 cl = hfsc_find_class(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
1190 if (cl == NULL || cl->level > 0)
1191 return NULL;
1192
1193 return cl;
1194}
1195
1196static int
1197hfsc_graft_class(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
10297b99 1198 struct Qdisc **old)
1da177e4
LT		 1199{
1200 struct hfsc_class *cl = (struct hfsc_class *)arg;
1201
1202 if (cl == NULL)
1203 return -ENOENT;
1204 if (cl->level > 0)
1205 return -EINVAL;
1206 if (new == NULL) {
5ce2d488 1207 new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
bb949fbd 1208 &pfifo_qdisc_ops,
be0d39d5 1209 cl->cl_common.classid);
1da177e4
LT		 1210 if (new == NULL)
1211 new = &noop_qdisc;
1212 }
1213
1214 sch_tree_lock(sch);
1215 hfsc_purge_queue(sch, cl);
b94c8afc
PM		 1216 *old = cl->qdisc;
1217 cl->qdisc = new;
1da177e4
LT		 1218 sch_tree_unlock(sch);
1219 return 0;
1220}
1221
1222static struct Qdisc *
1223hfsc_class_leaf(struct Qdisc *sch, unsigned long arg)
1224{
1225 struct hfsc_class *cl = (struct hfsc_class *)arg;
1226
1227 if (cl != NULL && cl->level == 0)
1228 return cl->qdisc;
1229
1230 return NULL;
1231}
1232
f973b913
PM		 1233static void
1234hfsc_qlen_notify(struct Qdisc *sch, unsigned long arg)
1235{
1236 struct hfsc_class *cl = (struct hfsc_class *)arg;
1237
1238 if (cl->qdisc->q.qlen == 0) {
1239 update_vf(cl, 0, 0);
1240 set_passive(cl);
1241 }
1242}
1243
1da177e4
LT		 1244static unsigned long
1245hfsc_get_class(struct Qdisc *sch, u32 classid)
1246{
1247 struct hfsc_class *cl = hfsc_find_class(classid, sch);
1248
1249 if (cl != NULL)
1250 cl->refcnt++;
1251
1252 return (unsigned long)cl;
1253}
1254
1255static void
1256hfsc_put_class(struct Qdisc *sch, unsigned long arg)
1257{
1258 struct hfsc_class *cl = (struct hfsc_class *)arg;
1259
1260 if (--cl->refcnt == 0)
1261 hfsc_destroy_class(sch, cl);
1262}
1263
1264static unsigned long
1265hfsc_bind_tcf(struct Qdisc *sch, unsigned long parent, u32 classid)
1266{
1267 struct hfsc_class *p = (struct hfsc_class *)parent;
1268 struct hfsc_class *cl = hfsc_find_class(classid, sch);
1269
1270 if (cl != NULL) {
1271 if (p != NULL && p->level <= cl->level)
1272 return 0;
1273 cl->filter_cnt++;
1274 }
1275
1276 return (unsigned long)cl;
1277}
1278
1279static void
1280hfsc_unbind_tcf(struct Qdisc *sch, unsigned long arg)
1281{
1282 struct hfsc_class *cl = (struct hfsc_class *)arg;
1283
1284 cl->filter_cnt--;
1285}
1286
1287static struct tcf_proto **
1288hfsc_tcf_chain(struct Qdisc *sch, unsigned long arg)
1289{
1290 struct hfsc_sched *q = qdisc_priv(sch);
1291 struct hfsc_class *cl = (struct hfsc_class *)arg;
1292
1293 if (cl == NULL)
1294 cl = &q->root;
1295
1296 return &cl->filter_list;
1297}
1298
1299static int
1300hfsc_dump_sc(struct sk_buff *skb, int attr, struct internal_sc *sc)
1301{
1302 struct tc_service_curve tsc;
1303
1304 tsc.m1 = sm2m(sc->sm1);
1305 tsc.d = dx2d(sc->dx);
1306 tsc.m2 = sm2m(sc->sm2);
1e90474c 1307 NLA_PUT(skb, attr, sizeof(tsc), &tsc);
1da177e4
LT		 1308
1309 return skb->len;
1310
1e90474c 1311 nla_put_failure:
1da177e4
LT		 1312 return -1;
1313}
1314
1315static inline int
1316hfsc_dump_curves(struct sk_buff *skb, struct hfsc_class *cl)
1317{
1318 if ((cl->cl_flags & HFSC_RSC) &&
1319 (hfsc_dump_sc(skb, TCA_HFSC_RSC, &cl->cl_rsc) < 0))
1e90474c 1320 goto nla_put_failure;
1da177e4
LT		 1321
1322 if ((cl->cl_flags & HFSC_FSC) &&
1323 (hfsc_dump_sc(skb, TCA_HFSC_FSC, &cl->cl_fsc) < 0))
1e90474c 1324 goto nla_put_failure;
1da177e4
LT		 1325
1326 if ((cl->cl_flags & HFSC_USC) &&
1327 (hfsc_dump_sc(skb, TCA_HFSC_USC, &cl->cl_usc) < 0))
1e90474c 1328 goto nla_put_failure;
1da177e4
LT		 1329
1330 return skb->len;
1331
1e90474c 1332 nla_put_failure:
1da177e4
LT		 1333 return -1;
1334}
1335
1336static int
1337hfsc_dump_class(struct Qdisc *sch, unsigned long arg, struct sk_buff *skb,
10297b99 1338 struct tcmsg *tcm)
1da177e4
LT		 1339{
1340 struct hfsc_class *cl = (struct hfsc_class *)arg;
4b3550ef 1341 struct nlattr *nest;
1da177e4 1342
be0d39d5
PM		 1343 tcm->tcm_parent = cl->cl_parent ? cl->cl_parent->cl_common.classid :
1344 TC_H_ROOT;
1345 tcm->tcm_handle = cl->cl_common.classid;
1da177e4
LT		 1346 if (cl->level == 0)
1347 tcm->tcm_info = cl->qdisc->handle;
1348
4b3550ef
PM		 1349 nest = nla_nest_start(skb, TCA_OPTIONS);
1350 if (nest == NULL)
1351 goto nla_put_failure;
1da177e4 1352 if (hfsc_dump_curves(skb, cl) < 0)
1e90474c 1353 goto nla_put_failure;
4b3550ef 1354 nla_nest_end(skb, nest);
1da177e4
LT		 1355 return skb->len;
1356
1e90474c 1357 nla_put_failure:
4b3550ef 1358 nla_nest_cancel(skb, nest);
bc3ed28c 1359 return -EMSGSIZE;
1da177e4
LT		 1360}
1361
1362static int
1363hfsc_dump_class_stats(struct Qdisc *sch, unsigned long arg,
1364 struct gnet_dump *d)
1365{
1366 struct hfsc_class *cl = (struct hfsc_class *)arg;
1367 struct tc_hfsc_stats xstats;
1368
1369 cl->qstats.qlen = cl->qdisc->q.qlen;
1370 xstats.level = cl->level;
1371 xstats.period = cl->cl_vtperiod;
1372 xstats.work = cl->cl_total;
1373 xstats.rtwork = cl->cl_cumul;
1374
1375 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
1da177e4 1376 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1da177e4
LT		 1377 gnet_stats_copy_queue(d, &cl->qstats) < 0)
1378 return -1;
1379
1380 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
1381}
1382
1383
1384
1385static void
1386hfsc_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1387{
1388 struct hfsc_sched *q = qdisc_priv(sch);
be0d39d5 1389 struct hlist_node *n;
1da177e4
LT		 1390 struct hfsc_class *cl;
1391 unsigned int i;
1392
1393 if (arg->stop)
1394 return;
1395
be0d39d5
PM		 1396 for (i = 0; i < q->clhash.hashsize; i++) {
1397 hlist_for_each_entry(cl, n, &q->clhash.hash[i],
1398 cl_common.hnode) {
1da177e4
LT		 1399 if (arg->count < arg->skip) {
1400 arg->count++;
1401 continue;
1402 }
1403 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1404 arg->stop = 1;
1405 return;
1406 }
1407 arg->count++;
1408 }
1409 }
1410}
1411
1412static void
ed2b229a 1413hfsc_schedule_watchdog(struct Qdisc *sch)
1da177e4
LT		 1414{
1415 struct hfsc_sched *q = qdisc_priv(sch);
1416 struct hfsc_class *cl;
1417 u64 next_time = 0;
1da177e4
LT		 1418
1419 if ((cl = eltree_get_minel(q)) != NULL)
1420 next_time = cl->cl_e;
1421 if (q->root.cl_cfmin != 0) {
1422 if (next_time == 0 || next_time > q->root.cl_cfmin)
1423 next_time = q->root.cl_cfmin;
1424 }
3d50f231 1425 WARN_ON(next_time == 0);
ed2b229a 1426 qdisc_watchdog_schedule(&q->watchdog, next_time);
1da177e4
LT		 1427}
1428
1429static int
1e90474c 1430hfsc_init_qdisc(struct Qdisc *sch, struct nlattr *opt)
1da177e4
LT		 1431{
1432 struct hfsc_sched *q = qdisc_priv(sch);
1433 struct tc_hfsc_qopt *qopt;
be0d39d5 1434 int err;
1da177e4 1435
1e90474c 1436 if (opt == NULL || nla_len(opt) < sizeof(*qopt))
1da177e4 1437 return -EINVAL;
1e90474c 1438 qopt = nla_data(opt);
1da177e4 1439
1da177e4 1440 q->defcls = qopt->defcls;
be0d39d5
PM		 1441 err = qdisc_class_hash_init(&q->clhash);
1442 if (err < 0)
1443 return err;
1da177e4
LT		 1444 q->eligible = RB_ROOT;
1445 INIT_LIST_HEAD(&q->droplist);
1da177e4 1446
be0d39d5 1447 q->root.cl_common.classid = sch->handle;
1da177e4 1448 q->root.refcnt = 1;
1da177e4 1449 q->root.sched = q;
5ce2d488 1450 q->root.qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
bb949fbd 1451 &pfifo_qdisc_ops,
9f9afec4 1452 sch->handle);
1da177e4
LT		 1453 if (q->root.qdisc == NULL)
1454 q->root.qdisc = &noop_qdisc;
1da177e4
LT		 1455 INIT_LIST_HEAD(&q->root.children);
1456 q->root.vt_tree = RB_ROOT;
1457 q->root.cf_tree = RB_ROOT;
1458
be0d39d5
PM		 1459 qdisc_class_hash_insert(&q->clhash, &q->root.cl_common);
1460 qdisc_class_hash_grow(sch, &q->clhash);
1da177e4 1461
ed2b229a 1462 qdisc_watchdog_init(&q->watchdog, sch);
1da177e4
LT		 1463
1464 return 0;
1465}
1466
1467static int
1e90474c 1468hfsc_change_qdisc(struct Qdisc *sch, struct nlattr *opt)
1da177e4
LT		 1469{
1470 struct hfsc_sched *q = qdisc_priv(sch);
1471 struct tc_hfsc_qopt *qopt;
1472
1e90474c 1473 if (opt == NULL || nla_len(opt) < sizeof(*qopt))
1da177e4 1474 return -EINVAL;
1e90474c 1475 qopt = nla_data(opt);
1da177e4
LT		 1476
1477 sch_tree_lock(sch);
1478 q->defcls = qopt->defcls;
1479 sch_tree_unlock(sch);
1480
1481 return 0;
1482}
1483
1484static void
1485hfsc_reset_class(struct hfsc_class *cl)
1486{
1487 cl->cl_total = 0;
1488 cl->cl_cumul = 0;
1489 cl->cl_d = 0;
1490 cl->cl_e = 0;
1491 cl->cl_vt = 0;
1492 cl->cl_vtadj = 0;
1493 cl->cl_vtoff = 0;
1494 cl->cl_cvtmin = 0;
1495 cl->cl_cvtmax = 0;
1496 cl->cl_cvtoff = 0;
1497 cl->cl_pcvtoff = 0;
1498 cl->cl_vtperiod = 0;
1499 cl->cl_parentperiod = 0;
1500 cl->cl_f = 0;
1501 cl->cl_myf = 0;
1502 cl->cl_myfadj = 0;
1503 cl->cl_cfmin = 0;
1504 cl->cl_nactive = 0;
1505
1506 cl->vt_tree = RB_ROOT;
1507 cl->cf_tree = RB_ROOT;
1508 qdisc_reset(cl->qdisc);
1509
1510 if (cl->cl_flags & HFSC_RSC)
1511 rtsc_init(&cl->cl_deadline, &cl->cl_rsc, 0, 0);
1512 if (cl->cl_flags & HFSC_FSC)
1513 rtsc_init(&cl->cl_virtual, &cl->cl_fsc, 0, 0);
1514 if (cl->cl_flags & HFSC_USC)
1515 rtsc_init(&cl->cl_ulimit, &cl->cl_usc, 0, 0);
1516}
1517
1518static void
1519hfsc_reset_qdisc(struct Qdisc *sch)
1520{
1521 struct hfsc_sched *q = qdisc_priv(sch);
1522 struct hfsc_class *cl;
be0d39d5 1523 struct hlist_node *n;
1da177e4
LT		 1524 unsigned int i;
1525
be0d39d5
PM		 1526 for (i = 0; i < q->clhash.hashsize; i++) {
1527 hlist_for_each_entry(cl, n, &q->clhash.hash[i], cl_common.hnode)
1da177e4
LT		 1528 hfsc_reset_class(cl);
1529 }
1da177e4
LT		 1530 q->eligible = RB_ROOT;
1531 INIT_LIST_HEAD(&q->droplist);
ed2b229a 1532 qdisc_watchdog_cancel(&q->watchdog);
1da177e4
LT		 1533 sch->q.qlen = 0;
1534}
1535
1536static void
1537hfsc_destroy_qdisc(struct Qdisc *sch)
1538{
1539 struct hfsc_sched *q = qdisc_priv(sch);
be0d39d5
PM		 1540 struct hlist_node *n, *next;
1541 struct hfsc_class *cl;
1da177e4
LT		 1542 unsigned int i;
1543
be0d39d5
PM		 1544 for (i = 0; i < q->clhash.hashsize; i++) {
1545 hlist_for_each_entry(cl, n, &q->clhash.hash[i], cl_common.hnode)
a4aebb83
PM		 1546 tcf_destroy_chain(&cl->filter_list);
1547 }
be0d39d5
PM		 1548 for (i = 0; i < q->clhash.hashsize; i++) {
1549 hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i],
1550 cl_common.hnode)
1da177e4
LT		 1551 hfsc_destroy_class(sch, cl);
1552 }
be0d39d5 1553 qdisc_class_hash_destroy(&q->clhash);
ed2b229a 1554 qdisc_watchdog_cancel(&q->watchdog);
1da177e4
LT		 1555}
1556
1557static int
1558hfsc_dump_qdisc(struct Qdisc *sch, struct sk_buff *skb)
1559{
1560 struct hfsc_sched *q = qdisc_priv(sch);
27a884dc 1561 unsigned char *b = skb_tail_pointer(skb);
1da177e4
LT		 1562 struct tc_hfsc_qopt qopt;
1563
1564 qopt.defcls = q->defcls;
1e90474c 1565 NLA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
1da177e4
LT		 1566 return skb->len;
1567
1e90474c 1568 nla_put_failure:
dc5fc579 1569 nlmsg_trim(skb, b);
1da177e4
LT		 1570 return -1;
1571}
1572
1573static int
1574hfsc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
1575{
1576 struct hfsc_class *cl;
dc0a0011 1577 int uninitialized_var(err);
1da177e4
LT		 1578
1579 cl = hfsc_classify(skb, sch, &err);
1580 if (cl == NULL) {
c27f339a 1581 if (err & __NET_XMIT_BYPASS)
1da177e4
LT		 1582 sch->qstats.drops++;
1583 kfree_skb(skb);
1584 return err;
1585 }
1586
5f86173b 1587 err = qdisc_enqueue(skb, cl->qdisc);
1da177e4 1588 if (unlikely(err != NET_XMIT_SUCCESS)) {
378a2f09
JP		 1589 if (net_xmit_drop_count(err)) {
1590 cl->qstats.drops++;
1591 sch->qstats.drops++;
1592 }
1da177e4
LT		 1593 return err;
1594 }
1595
1596 if (cl->qdisc->q.qlen == 1)
0abf77e5 1597 set_active(cl, qdisc_pkt_len(skb));
1da177e4
LT		 1598
1599 cl->bstats.packets++;
0abf77e5 1600 cl->bstats.bytes += qdisc_pkt_len(skb);
1da177e4 1601 sch->bstats.packets++;
0abf77e5 1602 sch->bstats.bytes += qdisc_pkt_len(skb);
1da177e4
LT		 1603 sch->q.qlen++;
1604
1605 return NET_XMIT_SUCCESS;
1606}
1607
1608static struct sk_buff *
1609hfsc_dequeue(struct Qdisc *sch)
1610{
1611 struct hfsc_sched *q = qdisc_priv(sch);
1612 struct hfsc_class *cl;
1613 struct sk_buff *skb;
1614 u64 cur_time;
1615 unsigned int next_len;
1616 int realtime = 0;
1617
1618 if (sch->q.qlen == 0)
1619 return NULL;
1da177e4 1620
3bebcda2 1621 cur_time = psched_get_time();
1da177e4
LT		 1622
1623 /*
1624 * if there are eligible classes, use real-time criteria.
1625 * find the class with the minimum deadline among
1626 * the eligible classes.
1627 */
1628 if ((cl = eltree_get_mindl(q, cur_time)) != NULL) {
1629 realtime = 1;
1630 } else {
1631 /*
1632 * use link-sharing criteria
1633 * get the class with the minimum vt in the hierarchy
1634 */
1635 cl = vttree_get_minvt(&q->root, cur_time);
1636 if (cl == NULL) {
1637 sch->qstats.overlimits++;
ed2b229a 1638 hfsc_schedule_watchdog(sch);
1da177e4
LT		 1639 return NULL;
1640 }
1641 }
1642
77be155c 1643 skb = qdisc_dequeue_peeked(cl->qdisc);
1da177e4
LT		 1644 if (skb == NULL) {
1645 if (net_ratelimit())
1646 printk("HFSC: Non-work-conserving qdisc ?\n");
1647 return NULL;
1648 }
1649
0abf77e5 1650 update_vf(cl, qdisc_pkt_len(skb), cur_time);
1da177e4 1651 if (realtime)
0abf77e5 1652 cl->cl_cumul += qdisc_pkt_len(skb);
1da177e4
LT		 1653
1654 if (cl->qdisc->q.qlen != 0) {
1655 if (cl->cl_flags & HFSC_RSC) {
1656 /* update ed */
1657 next_len = qdisc_peek_len(cl->qdisc);
1658 if (realtime)
1659 update_ed(cl, next_len);
1660 else
1661 update_d(cl, next_len);
1662 }
1663 } else {
1664 /* the class becomes passive */
1665 set_passive(cl);
1666 }
1667
1da177e4
LT		 1668 sch->flags &= ~TCQ_F_THROTTLED;
1669 sch->q.qlen--;
1670
1671 return skb;
1672}
1673
1da177e4
LT		 1674static unsigned int
1675hfsc_drop(struct Qdisc *sch)
1676{
1677 struct hfsc_sched *q = qdisc_priv(sch);
1678 struct hfsc_class *cl;
1679 unsigned int len;
1680
1681 list_for_each_entry(cl, &q->droplist, dlist) {
1682 if (cl->qdisc->ops->drop != NULL &&
1683 (len = cl->qdisc->ops->drop(cl->qdisc)) > 0) {
1684 if (cl->qdisc->q.qlen == 0) {
1685 update_vf(cl, 0, 0);
1686 set_passive(cl);
1687 } else {
1688 list_move_tail(&cl->dlist, &q->droplist);
1689 }
1690 cl->qstats.drops++;
1691 sch->qstats.drops++;
1692 sch->q.qlen--;
1693 return len;
1694 }
1695 }
1696 return 0;
1697}
1698
20fea08b 1699static const struct Qdisc_class_ops hfsc_class_ops = {
1da177e4
LT		 1700 .change = hfsc_change_class,
1701 .delete = hfsc_delete_class,
1702 .graft = hfsc_graft_class,
1703 .leaf = hfsc_class_leaf,
f973b913 1704 .qlen_notify = hfsc_qlen_notify,
1da177e4
LT		 1705 .get = hfsc_get_class,
1706 .put = hfsc_put_class,
1707 .bind_tcf = hfsc_bind_tcf,
1708 .unbind_tcf = hfsc_unbind_tcf,
1709 .tcf_chain = hfsc_tcf_chain,
1710 .dump = hfsc_dump_class,
1711 .dump_stats = hfsc_dump_class_stats,
1712 .walk = hfsc_walk
1713};
1714
20fea08b 1715static struct Qdisc_ops hfsc_qdisc_ops __read_mostly = {
1da177e4
LT		 1716 .id = "hfsc",
1717 .init = hfsc_init_qdisc,
1718 .change = hfsc_change_qdisc,
1719 .reset = hfsc_reset_qdisc,
1720 .destroy = hfsc_destroy_qdisc,
1721 .dump = hfsc_dump_qdisc,
1722 .enqueue = hfsc_enqueue,
1723 .dequeue = hfsc_dequeue,
77be155c 1724 .peek = qdisc_peek_dequeued,
1da177e4
LT		 1725 .drop = hfsc_drop,
1726 .cl_ops = &hfsc_class_ops,
1727 .priv_size = sizeof(struct hfsc_sched),
1728 .owner = THIS_MODULE
1729};
1730
1731static int __init
1732hfsc_init(void)
1733{
1734 return register_qdisc(&hfsc_qdisc_ops);
1735}
1736
1737static void __exit
1738hfsc_cleanup(void)
1739{
1740 unregister_qdisc(&hfsc_qdisc_ops);
1741}
1742
1743MODULE_LICENSE("GPL");
1744module_init(hfsc_init);
1745module_exit(hfsc_cleanup);
Linux 6.x block layer and io_uring tree(s)