1 // SPDX-License-Identifier: GPL-2.0-or-later <<
2 /* 1 /*
3 * net/sched/sch_htb.c Hierarchical token buc 2 * net/sched/sch_htb.c Hierarchical token bucket, feed tree version
4 * 3 *
>> 4 * This program is free software; you can redistribute it and/or
>> 5 * modify it under the terms of the GNU General Public License
>> 6 * as published by the Free Software Foundation; either version
>> 7 * 2 of the License, or (at your option) any later version.
>> 8 *
5 * Authors: Martin Devera, <devik@cdi.cz> 9 * Authors: Martin Devera, <devik@cdi.cz>
6 * 10 *
7 * Credits (in time order) for older HTB versi 11 * Credits (in time order) for older HTB versions:
8 * Stef Coene <stef.coene@docum.o 12 * Stef Coene <stef.coene@docum.org>
9 * HTB support at LARTC m 13 * HTB support at LARTC mailing list
10 * Ondrej Kraus, <krauso@barr.cz> 14 * Ondrej Kraus, <krauso@barr.cz>
11 * found missing INIT_QDI 15 * found missing INIT_QDISC(htb)
12 * Vladimir Smelhaus, Aamer Akhte 16 * Vladimir Smelhaus, Aamer Akhter, Bert Hubert
13 * helped a lot to locate 17 * helped a lot to locate nasty class stall bug
14 * Andi Kleen, Jamal Hadi, Bert H 18 * Andi Kleen, Jamal Hadi, Bert Hubert
15 * code review and helpfu 19 * code review and helpful comments on shaping
16 * Tomasz Wrona, <tw@eter.tym.pl> 20 * Tomasz Wrona, <tw@eter.tym.pl>
17 * created test case so t 21 * created test case so that I was able to fix nasty bug
18 * Wilfried Weissmann 22 * Wilfried Weissmann
19 * spotted bug in dequeue 23 * spotted bug in dequeue code and helped with fix
20 * Jiri Fojtasek 24 * Jiri Fojtasek
21 * fixed requeue routine 25 * fixed requeue routine
22 * and many others. thanks. 26 * and many others. thanks.
23 */ 27 */
24 #include <linux/module.h> 28 #include <linux/module.h>
25 #include <linux/moduleparam.h> 29 #include <linux/moduleparam.h>
26 #include <linux/types.h> 30 #include <linux/types.h>
27 #include <linux/kernel.h> 31 #include <linux/kernel.h>
28 #include <linux/string.h> 32 #include <linux/string.h>
29 #include <linux/errno.h> 33 #include <linux/errno.h>
30 #include <linux/skbuff.h> 34 #include <linux/skbuff.h>
31 #include <linux/list.h> 35 #include <linux/list.h>
32 #include <linux/compiler.h> 36 #include <linux/compiler.h>
33 #include <linux/rbtree.h> 37 #include <linux/rbtree.h>
34 #include <linux/workqueue.h> 38 #include <linux/workqueue.h>
35 #include <linux/slab.h> 39 #include <linux/slab.h>
36 #include <net/netlink.h> 40 #include <net/netlink.h>
37 #include <net/sch_generic.h> 41 #include <net/sch_generic.h>
38 #include <net/pkt_sched.h> 42 #include <net/pkt_sched.h>
39 #include <net/pkt_cls.h> 43 #include <net/pkt_cls.h>
40 44
41 /* HTB algorithm. 45 /* HTB algorithm.
42 Author: devik@cdi.cz 46 Author: devik@cdi.cz
43 ========================================== 47 ========================================================================
44 HTB is like TBF with multiple classes. It 48 HTB is like TBF with multiple classes. It is also similar to CBQ because
45 it allows to assign priority to each class 49 it allows to assign priority to each class in hierarchy.
46 In fact it is another implementation of Fl 50 In fact it is another implementation of Floyd's formal sharing.
47 51
48 Levels: 52 Levels:
49 Each class is assigned level. Leaf has ALW 53 Each class is assigned level. Leaf has ALWAYS level 0 and root
50 classes have level TC_HTB_MAXDEPTH-1. Inte 54 classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level
51 one less than their parent. 55 one less than their parent.
52 */ 56 */
53 57
54 static int htb_hysteresis __read_mostly = 0; / 58 static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */
55 #define HTB_VER 0x30011 /* major must !! 59 #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */
56 60
57 #if HTB_VER >> 16 != TC_HTB_PROTOVER 61 #if HTB_VER >> 16 != TC_HTB_PROTOVER
58 #error "Mismatched sch_htb.c and pkt_sch.h" 62 #error "Mismatched sch_htb.c and pkt_sch.h"
59 #endif 63 #endif
60 64
61 /* Module parameter and sysfs export */ 65 /* Module parameter and sysfs export */
62 module_param (htb_hysteresis, int, 0640); 66 module_param (htb_hysteresis, int, 0640);
63 MODULE_PARM_DESC(htb_hysteresis, "Hysteresis m 67 MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate");
64 68
65 static int htb_rate_est = 0; /* htb classes ha 69 static int htb_rate_est = 0; /* htb classes have a default rate estimator */
66 module_param(htb_rate_est, int, 0640); 70 module_param(htb_rate_est, int, 0640);
67 MODULE_PARM_DESC(htb_rate_est, "setup a defaul 71 MODULE_PARM_DESC(htb_rate_est, "setup a default rate estimator (4sec 16sec) for htb classes");
68 72
69 /* used internaly to keep status of single cla 73 /* used internaly to keep status of single class */
70 enum htb_cmode { 74 enum htb_cmode {
71 HTB_CANT_SEND, /* class can't 75 HTB_CANT_SEND, /* class can't send and can't borrow */
72 HTB_MAY_BORROW, /* class can't 76 HTB_MAY_BORROW, /* class can't send but may borrow */
73 HTB_CAN_SEND /* class can s 77 HTB_CAN_SEND /* class can send */
74 }; 78 };
75 79
76 struct htb_prio { 80 struct htb_prio {
77 union { 81 union {
78 struct rb_root row; 82 struct rb_root row;
79 struct rb_root feed; 83 struct rb_root feed;
80 }; 84 };
81 struct rb_node *ptr; 85 struct rb_node *ptr;
82 /* When class changes from state 1->2 86 /* When class changes from state 1->2 and disconnects from
83 * parent's feed then we lost ptr valu 87 * parent's feed then we lost ptr value and start from the
84 * first child again. Here we store cl 88 * first child again. Here we store classid of the
85 * last valid ptr (used when ptr is NU 89 * last valid ptr (used when ptr is NULL).
86 */ 90 */
87 u32 last_ptr_id; 91 u32 last_ptr_id;
88 }; 92 };
89 93
90 /* interior & leaf nodes; props specific to le 94 /* interior & leaf nodes; props specific to leaves are marked L:
91 * To reduce false sharing, place mostly read 95 * To reduce false sharing, place mostly read fields at beginning,
92 * and mostly written ones at the end. 96 * and mostly written ones at the end.
93 */ 97 */
94 struct htb_class { 98 struct htb_class {
95 struct Qdisc_class_common common; 99 struct Qdisc_class_common common;
96 struct psched_ratecfg rate; 100 struct psched_ratecfg rate;
97 struct psched_ratecfg ceil; 101 struct psched_ratecfg ceil;
98 s64 buffer, cbuffe 102 s64 buffer, cbuffer;/* token bucket depth/rate */
99 s64 mbuffer; 103 s64 mbuffer; /* max wait time */
100 u32 prio; 104 u32 prio; /* these two are used only by leaves... */
101 int quantum; 105 int quantum; /* but stored for parent-to-leaf return */
102 106
103 struct tcf_proto __rcu *filter_list; 107 struct tcf_proto __rcu *filter_list; /* class attached filters */
104 struct tcf_block *block; 108 struct tcf_block *block;
>> 109 int filter_cnt;
105 110
106 int level; 111 int level; /* our level (see above) */
107 unsigned int children; 112 unsigned int children;
108 struct htb_class *parent; 113 struct htb_class *parent; /* parent class */
109 114
110 struct net_rate_estimator __rcu *rate_ 115 struct net_rate_estimator __rcu *rate_est;
111 116
112 /* 117 /*
113 * Written often fields 118 * Written often fields
114 */ 119 */
115 struct gnet_stats_basic_sync bstats; !! 120 struct gnet_stats_basic_packed bstats;
116 struct gnet_stats_basic_sync bstats_bi <<
117 struct tc_htb_xstats xstats; /* our 121 struct tc_htb_xstats xstats; /* our special stats */
118 122
119 /* token bucket parameters */ 123 /* token bucket parameters */
120 s64 tokens, ctoken 124 s64 tokens, ctokens;/* current number of tokens */
121 s64 t_c; 125 s64 t_c; /* checkpoint time */
122 126
123 union { 127 union {
124 struct htb_class_leaf { 128 struct htb_class_leaf {
125 int defici 129 int deficit[TC_HTB_MAXDEPTH];
126 struct Qdisc *q; 130 struct Qdisc *q;
127 struct netdev_queue *o <<
128 } leaf; 131 } leaf;
129 struct htb_class_inner { 132 struct htb_class_inner {
130 struct htb_prio clprio 133 struct htb_prio clprio[TC_HTB_NUMPRIO];
131 } inner; 134 } inner;
132 }; 135 };
133 s64 pq_key; 136 s64 pq_key;
134 137
135 int prio_activity; 138 int prio_activity; /* for which prios are we active */
136 enum htb_cmode cmode; 139 enum htb_cmode cmode; /* current mode of the class */
137 struct rb_node pq_node; 140 struct rb_node pq_node; /* node for event queue */
138 struct rb_node node[TC_HTB_NU 141 struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */
139 142
140 unsigned int drops ____cacheline_align 143 unsigned int drops ____cacheline_aligned_in_smp;
141 unsigned int overlimits; 144 unsigned int overlimits;
142 }; 145 };
143 146
144 struct htb_level { 147 struct htb_level {
145 struct rb_root wait_pq; 148 struct rb_root wait_pq;
146 struct htb_prio hprio[TC_HTB_NUMPRIO]; 149 struct htb_prio hprio[TC_HTB_NUMPRIO];
147 }; 150 };
148 151
149 struct htb_sched { 152 struct htb_sched {
150 struct Qdisc_class_hash clhash; 153 struct Qdisc_class_hash clhash;
151 int defcls; 154 int defcls; /* class where unclassified flows go to */
152 int rate2quantum; 155 int rate2quantum; /* quant = rate / rate2quantum */
153 156
154 /* filters for qdisc itself */ 157 /* filters for qdisc itself */
155 struct tcf_proto __rcu *filter_list; 158 struct tcf_proto __rcu *filter_list;
156 struct tcf_block *block; 159 struct tcf_block *block;
157 160
158 #define HTB_WARN_TOOMANYEVENTS 0x1 161 #define HTB_WARN_TOOMANYEVENTS 0x1
159 unsigned int warned; /* onl 162 unsigned int warned; /* only one warning */
160 int direct_qlen; 163 int direct_qlen;
161 struct work_struct work; 164 struct work_struct work;
162 165
163 /* non shaped skbs; let them go direct 166 /* non shaped skbs; let them go directly thru */
164 struct qdisc_skb_head direct_queue; 167 struct qdisc_skb_head direct_queue;
165 u32 direct_pkts; !! 168 long direct_pkts;
166 u32 overlimits; <<
167 169
168 struct qdisc_watchdog watchdog; 170 struct qdisc_watchdog watchdog;
169 171
170 s64 now; /* cac 172 s64 now; /* cached dequeue time */
171 173
172 /* time of nearest event per level (ro 174 /* time of nearest event per level (row) */
173 s64 near_ev_cache[ 175 s64 near_ev_cache[TC_HTB_MAXDEPTH];
174 176
175 int row_mask[TC_HT 177 int row_mask[TC_HTB_MAXDEPTH];
176 178
177 struct htb_level hlevel[TC_HTB_ 179 struct htb_level hlevel[TC_HTB_MAXDEPTH];
178 <<
179 struct Qdisc **direct_qdisc <<
180 unsigned int num_direct_qdi <<
181 <<
182 bool offload; <<
183 }; 180 };
184 181
185 /* find class in global hash table using given 182 /* find class in global hash table using given handle */
186 static inline struct htb_class *htb_find(u32 h 183 static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch)
187 { 184 {
188 struct htb_sched *q = qdisc_priv(sch); 185 struct htb_sched *q = qdisc_priv(sch);
189 struct Qdisc_class_common *clc; 186 struct Qdisc_class_common *clc;
190 187
191 clc = qdisc_class_find(&q->clhash, han 188 clc = qdisc_class_find(&q->clhash, handle);
192 if (clc == NULL) 189 if (clc == NULL)
193 return NULL; 190 return NULL;
194 return container_of(clc, struct htb_cl 191 return container_of(clc, struct htb_class, common);
195 } 192 }
196 193
197 static unsigned long htb_search(struct Qdisc * 194 static unsigned long htb_search(struct Qdisc *sch, u32 handle)
198 { 195 {
199 return (unsigned long)htb_find(handle, 196 return (unsigned long)htb_find(handle, sch);
200 } 197 }
201 <<
202 #define HTB_DIRECT ((struct htb_class *)-1L) <<
203 <<
204 /** 198 /**
205 * htb_classify - classify a packet into class 199 * htb_classify - classify a packet into class
206 * @skb: the socket buffer <<
207 * @sch: the active queue discipline <<
208 * @qerr: pointer for returned status code <<
209 * 200 *
210 * It returns NULL if the packet should be dro 201 * It returns NULL if the packet should be dropped or -1 if the packet
211 * should be passed directly thru. In all othe 202 * should be passed directly thru. In all other cases leaf class is returned.
212 * We allow direct class selection by classid 203 * We allow direct class selection by classid in priority. The we examine
213 * filters in qdisc and in inner nodes (if hig 204 * filters in qdisc and in inner nodes (if higher filter points to the inner
214 * node). If we end up with classid MAJOR:0 we 205 * node). If we end up with classid MAJOR:0 we enqueue the skb into special
215 * internal fifo (direct). These packets then 206 * internal fifo (direct). These packets then go directly thru. If we still
216 * have no valid leaf we try to use MAJOR:defa 207 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful
217 * then finish and return direct queue. 208 * then finish and return direct queue.
218 */ 209 */
>> 210 #define HTB_DIRECT ((struct htb_class *)-1L)
>> 211
219 static struct htb_class *htb_classify(struct s 212 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch,
220 int *qer 213 int *qerr)
221 { 214 {
222 struct htb_sched *q = qdisc_priv(sch); 215 struct htb_sched *q = qdisc_priv(sch);
223 struct htb_class *cl; 216 struct htb_class *cl;
224 struct tcf_result res; 217 struct tcf_result res;
225 struct tcf_proto *tcf; 218 struct tcf_proto *tcf;
226 int result; 219 int result;
227 220
228 /* allow to select class by setting sk 221 /* allow to select class by setting skb->priority to valid classid;
229 * note that nfmark can be used too by 222 * note that nfmark can be used too by attaching filter fw with no
230 * rules in it 223 * rules in it
231 */ 224 */
232 if (skb->priority == sch->handle) 225 if (skb->priority == sch->handle)
233 return HTB_DIRECT; /* X:0 226 return HTB_DIRECT; /* X:0 (direct flow) selected */
234 cl = htb_find(skb->priority, sch); 227 cl = htb_find(skb->priority, sch);
235 if (cl) { 228 if (cl) {
236 if (cl->level == 0) 229 if (cl->level == 0)
237 return cl; 230 return cl;
238 /* Start with inner filter cha 231 /* Start with inner filter chain if a non-leaf class is selected */
239 tcf = rcu_dereference_bh(cl->f 232 tcf = rcu_dereference_bh(cl->filter_list);
240 } else { 233 } else {
241 tcf = rcu_dereference_bh(q->fi 234 tcf = rcu_dereference_bh(q->filter_list);
242 } 235 }
243 236
244 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_ 237 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
245 while (tcf && (result = tcf_classify(s !! 238 while (tcf && (result = tcf_classify(skb, tcf, &res, false)) >= 0) {
246 #ifdef CONFIG_NET_CLS_ACT 239 #ifdef CONFIG_NET_CLS_ACT
247 switch (result) { 240 switch (result) {
248 case TC_ACT_QUEUED: 241 case TC_ACT_QUEUED:
249 case TC_ACT_STOLEN: 242 case TC_ACT_STOLEN:
250 case TC_ACT_TRAP: 243 case TC_ACT_TRAP:
251 *qerr = NET_XMIT_SUCCE 244 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
252 fallthrough; !! 245 /* fall through */
253 case TC_ACT_SHOT: 246 case TC_ACT_SHOT:
254 return NULL; 247 return NULL;
255 } 248 }
256 #endif 249 #endif
257 cl = (void *)res.class; 250 cl = (void *)res.class;
258 if (!cl) { 251 if (!cl) {
259 if (res.classid == sch 252 if (res.classid == sch->handle)
260 return HTB_DIR 253 return HTB_DIRECT; /* X:0 (direct flow) */
261 cl = htb_find(res.clas 254 cl = htb_find(res.classid, sch);
262 if (!cl) 255 if (!cl)
263 break; /* fil 256 break; /* filter selected invalid classid */
264 } 257 }
265 if (!cl->level) 258 if (!cl->level)
266 return cl; /* we 259 return cl; /* we hit leaf; return it */
267 260
268 /* we have got inner class; ap 261 /* we have got inner class; apply inner filter chain */
269 tcf = rcu_dereference_bh(cl->f 262 tcf = rcu_dereference_bh(cl->filter_list);
270 } 263 }
271 /* classification failed; try to use d 264 /* classification failed; try to use default class */
272 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch-> 265 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
273 if (!cl || cl->level) 266 if (!cl || cl->level)
274 return HTB_DIRECT; /* bad 267 return HTB_DIRECT; /* bad default .. this is safe bet */
275 return cl; 268 return cl;
276 } 269 }
277 270
278 /** 271 /**
279 * htb_add_to_id_tree - adds class to the roun 272 * htb_add_to_id_tree - adds class to the round robin list
280 * @root: the root of the tree <<
281 * @cl: the class to add <<
282 * @prio: the give prio in class <<
283 * 273 *
284 * Routine adds class to the list (actually tr 274 * Routine adds class to the list (actually tree) sorted by classid.
285 * Make sure that class is not already on such 275 * Make sure that class is not already on such list for given prio.
286 */ 276 */
287 static void htb_add_to_id_tree(struct rb_root 277 static void htb_add_to_id_tree(struct rb_root *root,
288 struct htb_clas 278 struct htb_class *cl, int prio)
289 { 279 {
290 struct rb_node **p = &root->rb_node, * 280 struct rb_node **p = &root->rb_node, *parent = NULL;
291 281
292 while (*p) { 282 while (*p) {
293 struct htb_class *c; 283 struct htb_class *c;
294 parent = *p; 284 parent = *p;
295 c = rb_entry(parent, struct ht 285 c = rb_entry(parent, struct htb_class, node[prio]);
296 286
297 if (cl->common.classid > c->co 287 if (cl->common.classid > c->common.classid)
298 p = &parent->rb_right; 288 p = &parent->rb_right;
299 else 289 else
300 p = &parent->rb_left; 290 p = &parent->rb_left;
301 } 291 }
302 rb_link_node(&cl->node[prio], parent, 292 rb_link_node(&cl->node[prio], parent, p);
303 rb_insert_color(&cl->node[prio], root) 293 rb_insert_color(&cl->node[prio], root);
304 } 294 }
305 295
306 /** 296 /**
307 * htb_add_to_wait_tree - adds class to the ev 297 * htb_add_to_wait_tree - adds class to the event queue with delay
308 * @q: the priority event queue <<
309 * @cl: the class to add <<
310 * @delay: delay in microseconds <<
311 * 298 *
312 * The class is added to priority event queue 299 * The class is added to priority event queue to indicate that class will
313 * change its mode in cl->pq_key microseconds. 300 * change its mode in cl->pq_key microseconds. Make sure that class is not
314 * already in the queue. 301 * already in the queue.
315 */ 302 */
316 static void htb_add_to_wait_tree(struct htb_sc 303 static void htb_add_to_wait_tree(struct htb_sched *q,
317 struct htb_cl 304 struct htb_class *cl, s64 delay)
318 { 305 {
319 struct rb_node **p = &q->hlevel[cl->le 306 struct rb_node **p = &q->hlevel[cl->level].wait_pq.rb_node, *parent = NULL;
320 307
321 cl->pq_key = q->now + delay; 308 cl->pq_key = q->now + delay;
322 if (cl->pq_key == q->now) 309 if (cl->pq_key == q->now)
323 cl->pq_key++; 310 cl->pq_key++;
324 311
325 /* update the nearest event cache */ 312 /* update the nearest event cache */
326 if (q->near_ev_cache[cl->level] > cl-> 313 if (q->near_ev_cache[cl->level] > cl->pq_key)
327 q->near_ev_cache[cl->level] = 314 q->near_ev_cache[cl->level] = cl->pq_key;
328 315
329 while (*p) { 316 while (*p) {
330 struct htb_class *c; 317 struct htb_class *c;
331 parent = *p; 318 parent = *p;
332 c = rb_entry(parent, struct ht 319 c = rb_entry(parent, struct htb_class, pq_node);
333 if (cl->pq_key >= c->pq_key) 320 if (cl->pq_key >= c->pq_key)
334 p = &parent->rb_right; 321 p = &parent->rb_right;
335 else 322 else
336 p = &parent->rb_left; 323 p = &parent->rb_left;
337 } 324 }
338 rb_link_node(&cl->pq_node, parent, p); 325 rb_link_node(&cl->pq_node, parent, p);
339 rb_insert_color(&cl->pq_node, &q->hlev 326 rb_insert_color(&cl->pq_node, &q->hlevel[cl->level].wait_pq);
340 } 327 }
341 328
342 /** 329 /**
343 * htb_next_rb_node - finds next node in binar 330 * htb_next_rb_node - finds next node in binary tree
344 * @n: the current node in binary tree <<
345 * 331 *
346 * When we are past last key we return NULL. 332 * When we are past last key we return NULL.
347 * Average complexity is 2 steps per call. 333 * Average complexity is 2 steps per call.
348 */ 334 */
349 static inline void htb_next_rb_node(struct rb_ 335 static inline void htb_next_rb_node(struct rb_node **n)
350 { 336 {
351 *n = rb_next(*n); 337 *n = rb_next(*n);
352 } 338 }
353 339
354 /** 340 /**
355 * htb_add_class_to_row - add class to its row 341 * htb_add_class_to_row - add class to its row
356 * @q: the priority event queue <<
357 * @cl: the class to add <<
358 * @mask: the given priorities in class in bit <<
359 * 342 *
360 * The class is added to row at priorities mar 343 * The class is added to row at priorities marked in mask.
361 * It does nothing if mask == 0. 344 * It does nothing if mask == 0.
362 */ 345 */
363 static inline void htb_add_class_to_row(struct 346 static inline void htb_add_class_to_row(struct htb_sched *q,
364 struct 347 struct htb_class *cl, int mask)
365 { 348 {
366 q->row_mask[cl->level] |= mask; 349 q->row_mask[cl->level] |= mask;
367 while (mask) { 350 while (mask) {
368 int prio = ffz(~mask); 351 int prio = ffz(~mask);
369 mask &= ~(1 << prio); 352 mask &= ~(1 << prio);
370 htb_add_to_id_tree(&q->hlevel[ 353 htb_add_to_id_tree(&q->hlevel[cl->level].hprio[prio].row, cl, prio);
371 } 354 }
372 } 355 }
373 356
374 /* If this triggers, it is a bug in this code, 357 /* If this triggers, it is a bug in this code, but it need not be fatal */
375 static void htb_safe_rb_erase(struct rb_node * 358 static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root)
376 { 359 {
377 if (RB_EMPTY_NODE(rb)) { 360 if (RB_EMPTY_NODE(rb)) {
378 WARN_ON(1); 361 WARN_ON(1);
379 } else { 362 } else {
380 rb_erase(rb, root); 363 rb_erase(rb, root);
381 RB_CLEAR_NODE(rb); 364 RB_CLEAR_NODE(rb);
382 } 365 }
383 } 366 }
384 367
385 368
386 /** 369 /**
387 * htb_remove_class_from_row - removes class f 370 * htb_remove_class_from_row - removes class from its row
388 * @q: the priority event queue <<
389 * @cl: the class to add <<
390 * @mask: the given priorities in class in bit <<
391 * 371 *
392 * The class is removed from row at priorities 372 * The class is removed from row at priorities marked in mask.
393 * It does nothing if mask == 0. 373 * It does nothing if mask == 0.
394 */ 374 */
395 static inline void htb_remove_class_from_row(s 375 static inline void htb_remove_class_from_row(struct htb_sched *q,
396 376 struct htb_class *cl, int mask)
397 { 377 {
398 int m = 0; 378 int m = 0;
399 struct htb_level *hlevel = &q->hlevel[ 379 struct htb_level *hlevel = &q->hlevel[cl->level];
400 380
401 while (mask) { 381 while (mask) {
402 int prio = ffz(~mask); 382 int prio = ffz(~mask);
403 struct htb_prio *hprio = &hlev 383 struct htb_prio *hprio = &hlevel->hprio[prio];
404 384
405 mask &= ~(1 << prio); 385 mask &= ~(1 << prio);
406 if (hprio->ptr == cl->node + p 386 if (hprio->ptr == cl->node + prio)
407 htb_next_rb_node(&hpri 387 htb_next_rb_node(&hprio->ptr);
408 388
409 htb_safe_rb_erase(cl->node + p 389 htb_safe_rb_erase(cl->node + prio, &hprio->row);
410 if (!hprio->row.rb_node) 390 if (!hprio->row.rb_node)
411 m |= 1 << prio; 391 m |= 1 << prio;
412 } 392 }
413 q->row_mask[cl->level] &= ~m; 393 q->row_mask[cl->level] &= ~m;
414 } 394 }
415 395
416 /** 396 /**
417 * htb_activate_prios - creates active classe' 397 * htb_activate_prios - creates active classe's feed chain
418 * @q: the priority event queue <<
419 * @cl: the class to activate <<
420 * 398 *
421 * The class is connected to ancestors and/or 399 * The class is connected to ancestors and/or appropriate rows
422 * for priorities it is participating on. cl-> 400 * for priorities it is participating on. cl->cmode must be new
423 * (activated) mode. It does nothing if cl->pr 401 * (activated) mode. It does nothing if cl->prio_activity == 0.
424 */ 402 */
425 static void htb_activate_prios(struct htb_sche 403 static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl)
426 { 404 {
427 struct htb_class *p = cl->parent; 405 struct htb_class *p = cl->parent;
428 long m, mask = cl->prio_activity; 406 long m, mask = cl->prio_activity;
429 407
430 while (cl->cmode == HTB_MAY_BORROW && 408 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
431 m = mask; 409 m = mask;
432 while (m) { 410 while (m) {
433 unsigned int prio = ff !! 411 int prio = ffz(~m);
434 <<
435 if (WARN_ON_ONCE(prio <<
436 break; <<
437 m &= ~(1 << prio); 412 m &= ~(1 << prio);
438 413
439 if (p->inner.clprio[pr 414 if (p->inner.clprio[prio].feed.rb_node)
440 /* parent alre 415 /* parent already has its feed in use so that
441 * reset bit i 416 * reset bit in mask as parent is already ok
442 */ 417 */
443 mask &= ~(1 << 418 mask &= ~(1 << prio);
444 419
445 htb_add_to_id_tree(&p- 420 htb_add_to_id_tree(&p->inner.clprio[prio].feed, cl, prio);
446 } 421 }
447 p->prio_activity |= mask; 422 p->prio_activity |= mask;
448 cl = p; 423 cl = p;
449 p = cl->parent; 424 p = cl->parent;
450 425
451 } 426 }
452 if (cl->cmode == HTB_CAN_SEND && mask) 427 if (cl->cmode == HTB_CAN_SEND && mask)
453 htb_add_class_to_row(q, cl, ma 428 htb_add_class_to_row(q, cl, mask);
454 } 429 }
455 430
456 /** 431 /**
457 * htb_deactivate_prios - remove class from fe 432 * htb_deactivate_prios - remove class from feed chain
458 * @q: the priority event queue <<
459 * @cl: the class to deactivate <<
460 * 433 *
461 * cl->cmode must represent old mode (before d 434 * cl->cmode must represent old mode (before deactivation). It does
462 * nothing if cl->prio_activity == 0. Class is 435 * nothing if cl->prio_activity == 0. Class is removed from all feed
463 * chains and rows. 436 * chains and rows.
464 */ 437 */
465 static void htb_deactivate_prios(struct htb_sc 438 static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl)
466 { 439 {
467 struct htb_class *p = cl->parent; 440 struct htb_class *p = cl->parent;
468 long m, mask = cl->prio_activity; 441 long m, mask = cl->prio_activity;
469 442
470 while (cl->cmode == HTB_MAY_BORROW && 443 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
471 m = mask; 444 m = mask;
472 mask = 0; 445 mask = 0;
473 while (m) { 446 while (m) {
474 int prio = ffz(~m); 447 int prio = ffz(~m);
475 m &= ~(1 << prio); 448 m &= ~(1 << prio);
476 449
477 if (p->inner.clprio[pr 450 if (p->inner.clprio[prio].ptr == cl->node + prio) {
478 /* we are remo 451 /* we are removing child which is pointed to from
479 * parent feed 452 * parent feed - forget the pointer but remember
480 * classid 453 * classid
481 */ 454 */
482 p->inner.clpri 455 p->inner.clprio[prio].last_ptr_id = cl->common.classid;
483 p->inner.clpri 456 p->inner.clprio[prio].ptr = NULL;
484 } 457 }
485 458
486 htb_safe_rb_erase(cl-> 459 htb_safe_rb_erase(cl->node + prio,
487 &p-> 460 &p->inner.clprio[prio].feed);
488 461
489 if (!p->inner.clprio[p 462 if (!p->inner.clprio[prio].feed.rb_node)
490 mask |= 1 << p 463 mask |= 1 << prio;
491 } 464 }
492 465
493 p->prio_activity &= ~mask; 466 p->prio_activity &= ~mask;
494 cl = p; 467 cl = p;
495 p = cl->parent; 468 p = cl->parent;
496 469
497 } 470 }
498 if (cl->cmode == HTB_CAN_SEND && mask) 471 if (cl->cmode == HTB_CAN_SEND && mask)
499 htb_remove_class_from_row(q, c 472 htb_remove_class_from_row(q, cl, mask);
500 } 473 }
501 474
502 static inline s64 htb_lowater(const struct htb 475 static inline s64 htb_lowater(const struct htb_class *cl)
503 { 476 {
504 if (htb_hysteresis) 477 if (htb_hysteresis)
505 return cl->cmode != HTB_CANT_S 478 return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0;
506 else 479 else
507 return 0; 480 return 0;
508 } 481 }
509 static inline s64 htb_hiwater(const struct htb 482 static inline s64 htb_hiwater(const struct htb_class *cl)
510 { 483 {
511 if (htb_hysteresis) 484 if (htb_hysteresis)
512 return cl->cmode == HTB_CAN_SE 485 return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0;
513 else 486 else
514 return 0; 487 return 0;
515 } 488 }
516 489
517 490
518 /** 491 /**
519 * htb_class_mode - computes and returns curre 492 * htb_class_mode - computes and returns current class mode
520 * @cl: the target class <<
521 * @diff: diff time in microseconds <<
522 * 493 *
523 * It computes cl's mode at time cl->t_c+diff 494 * It computes cl's mode at time cl->t_c+diff and returns it. If mode
524 * is not HTB_CAN_SEND then cl->pq_key is upda 495 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
525 * from now to time when cl will change its st 496 * from now to time when cl will change its state.
526 * Also it is worth to note that class mode do 497 * Also it is worth to note that class mode doesn't change simply
527 * at cl->{c,}tokens == 0 but there can rather 498 * at cl->{c,}tokens == 0 but there can rather be hysteresis of
528 * 0 .. -cl->{c,}buffer range. It is meant to 499 * 0 .. -cl->{c,}buffer range. It is meant to limit number of
529 * mode transitions per time unit. The speed g 500 * mode transitions per time unit. The speed gain is about 1/6.
530 */ 501 */
531 static inline enum htb_cmode 502 static inline enum htb_cmode
532 htb_class_mode(struct htb_class *cl, s64 *diff 503 htb_class_mode(struct htb_class *cl, s64 *diff)
533 { 504 {
534 s64 toks; 505 s64 toks;
535 506
536 if ((toks = (cl->ctokens + *diff)) < h 507 if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) {
537 *diff = -toks; 508 *diff = -toks;
538 return HTB_CANT_SEND; 509 return HTB_CANT_SEND;
539 } 510 }
540 511
541 if ((toks = (cl->tokens + *diff)) >= h 512 if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl))
542 return HTB_CAN_SEND; 513 return HTB_CAN_SEND;
543 514
544 *diff = -toks; 515 *diff = -toks;
545 return HTB_MAY_BORROW; 516 return HTB_MAY_BORROW;
546 } 517 }
547 518
548 /** 519 /**
549 * htb_change_class_mode - changes classe's mo 520 * htb_change_class_mode - changes classe's mode
550 * @q: the priority event queue <<
551 * @cl: the target class <<
552 * @diff: diff time in microseconds <<
553 * 521 *
554 * This should be the only way how to change c 522 * This should be the only way how to change classe's mode under normal
555 * circumstances. Routine will update feed lis !! 523 * cirsumstances. Routine will update feed lists linkage, change mode
556 * and add class to the wait event queue if ap 524 * and add class to the wait event queue if appropriate. New mode should
557 * be different from old one and cl->pq_key ha 525 * be different from old one and cl->pq_key has to be valid if changing
558 * to mode other than HTB_CAN_SEND (see htb_ad 526 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
559 */ 527 */
560 static void 528 static void
561 htb_change_class_mode(struct htb_sched *q, str 529 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, s64 *diff)
562 { 530 {
563 enum htb_cmode new_mode = htb_class_mo 531 enum htb_cmode new_mode = htb_class_mode(cl, diff);
564 532
565 if (new_mode == cl->cmode) 533 if (new_mode == cl->cmode)
566 return; 534 return;
567 535
568 if (new_mode == HTB_CANT_SEND) { !! 536 if (new_mode == HTB_CANT_SEND)
569 cl->overlimits++; 537 cl->overlimits++;
570 q->overlimits++; <<
571 } <<
572 538
573 if (cl->prio_activity) { /* not 539 if (cl->prio_activity) { /* not necessary: speed optimization */
574 if (cl->cmode != HTB_CANT_SEND 540 if (cl->cmode != HTB_CANT_SEND)
575 htb_deactivate_prios(q 541 htb_deactivate_prios(q, cl);
576 cl->cmode = new_mode; 542 cl->cmode = new_mode;
577 if (new_mode != HTB_CANT_SEND) 543 if (new_mode != HTB_CANT_SEND)
578 htb_activate_prios(q, 544 htb_activate_prios(q, cl);
579 } else 545 } else
580 cl->cmode = new_mode; 546 cl->cmode = new_mode;
581 } 547 }
582 548
583 /** 549 /**
584 * htb_activate - inserts leaf cl into appropr 550 * htb_activate - inserts leaf cl into appropriate active feeds
585 * @q: the priority event queue <<
586 * @cl: the target class <<
587 * 551 *
588 * Routine learns (new) priority of leaf and a 552 * Routine learns (new) priority of leaf and activates feed chain
589 * for the prio. It can be called on already a 553 * for the prio. It can be called on already active leaf safely.
590 * It also adds leaf into droplist. 554 * It also adds leaf into droplist.
591 */ 555 */
592 static inline void htb_activate(struct htb_sch 556 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl)
593 { 557 {
594 WARN_ON(cl->level || !cl->leaf.q || !c 558 WARN_ON(cl->level || !cl->leaf.q || !cl->leaf.q->q.qlen);
595 559
596 if (!cl->prio_activity) { 560 if (!cl->prio_activity) {
597 cl->prio_activity = 1 << cl->p 561 cl->prio_activity = 1 << cl->prio;
598 htb_activate_prios(q, cl); 562 htb_activate_prios(q, cl);
599 } 563 }
600 } 564 }
601 565
602 /** 566 /**
603 * htb_deactivate - remove leaf cl from active 567 * htb_deactivate - remove leaf cl from active feeds
604 * @q: the priority event queue <<
605 * @cl: the target class <<
606 * 568 *
607 * Make sure that leaf is active. In the other 569 * Make sure that leaf is active. In the other words it can't be called
608 * with non-active leaf. It also removes class 570 * with non-active leaf. It also removes class from the drop list.
609 */ 571 */
610 static inline void htb_deactivate(struct htb_s 572 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl)
611 { 573 {
612 WARN_ON(!cl->prio_activity); 574 WARN_ON(!cl->prio_activity);
613 575
614 htb_deactivate_prios(q, cl); 576 htb_deactivate_prios(q, cl);
615 cl->prio_activity = 0; 577 cl->prio_activity = 0;
616 } 578 }
617 579
618 static int htb_enqueue(struct sk_buff *skb, st 580 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch,
619 struct sk_buff **to_fre 581 struct sk_buff **to_free)
620 { 582 {
621 int ret; !! 583 int uninitialized_var(ret);
622 unsigned int len = qdisc_pkt_len(skb); 584 unsigned int len = qdisc_pkt_len(skb);
623 struct htb_sched *q = qdisc_priv(sch); 585 struct htb_sched *q = qdisc_priv(sch);
624 struct htb_class *cl = htb_classify(sk 586 struct htb_class *cl = htb_classify(skb, sch, &ret);
625 587
626 if (cl == HTB_DIRECT) { 588 if (cl == HTB_DIRECT) {
627 /* enqueue to helper queue */ 589 /* enqueue to helper queue */
628 if (q->direct_queue.qlen < q-> 590 if (q->direct_queue.qlen < q->direct_qlen) {
629 __qdisc_enqueue_tail(s 591 __qdisc_enqueue_tail(skb, &q->direct_queue);
630 q->direct_pkts++; 592 q->direct_pkts++;
631 } else { 593 } else {
632 return qdisc_drop(skb, 594 return qdisc_drop(skb, sch, to_free);
633 } 595 }
634 #ifdef CONFIG_NET_CLS_ACT 596 #ifdef CONFIG_NET_CLS_ACT
635 } else if (!cl) { 597 } else if (!cl) {
636 if (ret & __NET_XMIT_BYPASS) 598 if (ret & __NET_XMIT_BYPASS)
637 qdisc_qstats_drop(sch) 599 qdisc_qstats_drop(sch);
638 __qdisc_drop(skb, to_free); 600 __qdisc_drop(skb, to_free);
639 return ret; 601 return ret;
640 #endif 602 #endif
641 } else if ((ret = qdisc_enqueue(skb, c 603 } else if ((ret = qdisc_enqueue(skb, cl->leaf.q,
642 to_fre 604 to_free)) != NET_XMIT_SUCCESS) {
643 if (net_xmit_drop_count(ret)) 605 if (net_xmit_drop_count(ret)) {
644 qdisc_qstats_drop(sch) 606 qdisc_qstats_drop(sch);
645 cl->drops++; 607 cl->drops++;
646 } 608 }
647 return ret; 609 return ret;
648 } else { 610 } else {
649 htb_activate(q, cl); 611 htb_activate(q, cl);
650 } 612 }
651 613
652 sch->qstats.backlog += len; 614 sch->qstats.backlog += len;
653 sch->q.qlen++; 615 sch->q.qlen++;
654 return NET_XMIT_SUCCESS; 616 return NET_XMIT_SUCCESS;
655 } 617 }
656 618
657 static inline void htb_accnt_tokens(struct htb 619 static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, s64 diff)
658 { 620 {
659 s64 toks = diff + cl->tokens; 621 s64 toks = diff + cl->tokens;
660 622
661 if (toks > cl->buffer) 623 if (toks > cl->buffer)
662 toks = cl->buffer; 624 toks = cl->buffer;
663 toks -= (s64) psched_l2t_ns(&cl->rate, 625 toks -= (s64) psched_l2t_ns(&cl->rate, bytes);
664 if (toks <= -cl->mbuffer) 626 if (toks <= -cl->mbuffer)
665 toks = 1 - cl->mbuffer; 627 toks = 1 - cl->mbuffer;
666 628
667 cl->tokens = toks; 629 cl->tokens = toks;
668 } 630 }
669 631
670 static inline void htb_accnt_ctokens(struct ht 632 static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, s64 diff)
671 { 633 {
672 s64 toks = diff + cl->ctokens; 634 s64 toks = diff + cl->ctokens;
673 635
674 if (toks > cl->cbuffer) 636 if (toks > cl->cbuffer)
675 toks = cl->cbuffer; 637 toks = cl->cbuffer;
676 toks -= (s64) psched_l2t_ns(&cl->ceil, 638 toks -= (s64) psched_l2t_ns(&cl->ceil, bytes);
677 if (toks <= -cl->mbuffer) 639 if (toks <= -cl->mbuffer)
678 toks = 1 - cl->mbuffer; 640 toks = 1 - cl->mbuffer;
679 641
680 cl->ctokens = toks; 642 cl->ctokens = toks;
681 } 643 }
682 644
683 /** 645 /**
684 * htb_charge_class - charges amount "bytes" t 646 * htb_charge_class - charges amount "bytes" to leaf and ancestors
685 * @q: the priority event queue <<
686 * @cl: the class to start iterate <<
687 * @level: the minimum level to account <<
688 * @skb: the socket buffer <<
689 * 647 *
690 * Routine assumes that packet "bytes" long wa 648 * Routine assumes that packet "bytes" long was dequeued from leaf cl
691 * borrowing from "level". It accounts bytes t 649 * borrowing from "level". It accounts bytes to ceil leaky bucket for
692 * leaf and all ancestors and to rate bucket f 650 * leaf and all ancestors and to rate bucket for ancestors at levels
693 * "level" and higher. It also handles possibl 651 * "level" and higher. It also handles possible change of mode resulting
694 * from the update. Note that mode can also in 652 * from the update. Note that mode can also increase here (MAY_BORROW to
695 * CAN_SEND) because we can use more precise c 653 * CAN_SEND) because we can use more precise clock that event queue here.
696 * In such case we remove class from event que 654 * In such case we remove class from event queue first.
697 */ 655 */
698 static void htb_charge_class(struct htb_sched 656 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl,
699 int level, struct 657 int level, struct sk_buff *skb)
700 { 658 {
701 int bytes = qdisc_pkt_len(skb); 659 int bytes = qdisc_pkt_len(skb);
702 enum htb_cmode old_mode; 660 enum htb_cmode old_mode;
703 s64 diff; 661 s64 diff;
704 662
705 while (cl) { 663 while (cl) {
706 diff = min_t(s64, q->now - cl- 664 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer);
707 if (cl->level >= level) { 665 if (cl->level >= level) {
708 if (cl->level == level 666 if (cl->level == level)
709 cl->xstats.len 667 cl->xstats.lends++;
710 htb_accnt_tokens(cl, b 668 htb_accnt_tokens(cl, bytes, diff);
711 } else { 669 } else {
712 cl->xstats.borrows++; 670 cl->xstats.borrows++;
713 cl->tokens += diff; 671 cl->tokens += diff; /* we moved t_c; update tokens */
714 } 672 }
715 htb_accnt_ctokens(cl, bytes, d 673 htb_accnt_ctokens(cl, bytes, diff);
716 cl->t_c = q->now; 674 cl->t_c = q->now;
717 675
718 old_mode = cl->cmode; 676 old_mode = cl->cmode;
719 diff = 0; 677 diff = 0;
720 htb_change_class_mode(q, cl, & 678 htb_change_class_mode(q, cl, &diff);
721 if (old_mode != cl->cmode) { 679 if (old_mode != cl->cmode) {
722 if (old_mode != HTB_CA 680 if (old_mode != HTB_CAN_SEND)
723 htb_safe_rb_er 681 htb_safe_rb_erase(&cl->pq_node, &q->hlevel[cl->level].wait_pq);
724 if (cl->cmode != HTB_C 682 if (cl->cmode != HTB_CAN_SEND)
725 htb_add_to_wai 683 htb_add_to_wait_tree(q, cl, diff);
726 } 684 }
727 685
728 /* update basic stats except f 686 /* update basic stats except for leaves which are already updated */
729 if (cl->level) 687 if (cl->level)
730 bstats_update(&cl->bst 688 bstats_update(&cl->bstats, skb);
731 689
732 cl = cl->parent; 690 cl = cl->parent;
733 } 691 }
734 } 692 }
735 693
736 /** 694 /**
737 * htb_do_events - make mode changes to classe 695 * htb_do_events - make mode changes to classes at the level
738 * @q: the priority event queue <<
739 * @level: which wait_pq in 'q->hlevel' <<
740 * @start: start jiffies <<
741 * 696 *
742 * Scans event queue for pending events and ap 697 * Scans event queue for pending events and applies them. Returns time of
743 * next pending event (0 for no event in pq, q 698 * next pending event (0 for no event in pq, q->now for too many events).
744 * Note: Applied are events whose have cl->pq_ 699 * Note: Applied are events whose have cl->pq_key <= q->now.
745 */ 700 */
746 static s64 htb_do_events(struct htb_sched *q, 701 static s64 htb_do_events(struct htb_sched *q, const int level,
747 unsigned long start) 702 unsigned long start)
748 { 703 {
749 /* don't run for longer than 2 jiffies 704 /* don't run for longer than 2 jiffies; 2 is used instead of
750 * 1 to simplify things when jiffy is 705 * 1 to simplify things when jiffy is going to be incremented
751 * too soon 706 * too soon
752 */ 707 */
753 unsigned long stop_at = start + 2; 708 unsigned long stop_at = start + 2;
754 struct rb_root *wait_pq = &q->hlevel[l 709 struct rb_root *wait_pq = &q->hlevel[level].wait_pq;
755 710
756 while (time_before(jiffies, stop_at)) 711 while (time_before(jiffies, stop_at)) {
757 struct htb_class *cl; 712 struct htb_class *cl;
758 s64 diff; 713 s64 diff;
759 struct rb_node *p = rb_first(w 714 struct rb_node *p = rb_first(wait_pq);
760 715
761 if (!p) 716 if (!p)
762 return 0; 717 return 0;
763 718
764 cl = rb_entry(p, struct htb_cl 719 cl = rb_entry(p, struct htb_class, pq_node);
765 if (cl->pq_key > q->now) 720 if (cl->pq_key > q->now)
766 return cl->pq_key; 721 return cl->pq_key;
767 722
768 htb_safe_rb_erase(p, wait_pq); 723 htb_safe_rb_erase(p, wait_pq);
769 diff = min_t(s64, q->now - cl- 724 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer);
770 htb_change_class_mode(q, cl, & 725 htb_change_class_mode(q, cl, &diff);
771 if (cl->cmode != HTB_CAN_SEND) 726 if (cl->cmode != HTB_CAN_SEND)
772 htb_add_to_wait_tree(q 727 htb_add_to_wait_tree(q, cl, diff);
773 } 728 }
774 729
775 /* too much load - let's continue afte 730 /* too much load - let's continue after a break for scheduling */
776 if (!(q->warned & HTB_WARN_TOOMANYEVEN 731 if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) {
777 pr_warn("htb: too many events! 732 pr_warn("htb: too many events!\n");
778 q->warned |= HTB_WARN_TOOMANYE 733 q->warned |= HTB_WARN_TOOMANYEVENTS;
779 } 734 }
780 735
781 return q->now; 736 return q->now;
782 } 737 }
783 738
784 /* Returns class->node+prio from id-tree where 739 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
785 * is no such one exists. 740 * is no such one exists.
786 */ 741 */
787 static struct rb_node *htb_id_find_next_upper( 742 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n,
788 743 u32 id)
789 { 744 {
790 struct rb_node *r = NULL; 745 struct rb_node *r = NULL;
791 while (n) { 746 while (n) {
792 struct htb_class *cl = 747 struct htb_class *cl =
793 rb_entry(n, struct htb_cla 748 rb_entry(n, struct htb_class, node[prio]);
794 749
795 if (id > cl->common.classid) { 750 if (id > cl->common.classid) {
796 n = n->rb_right; 751 n = n->rb_right;
797 } else if (id < cl->common.cla 752 } else if (id < cl->common.classid) {
798 r = n; 753 r = n;
799 n = n->rb_left; 754 n = n->rb_left;
800 } else { 755 } else {
801 return n; 756 return n;
802 } 757 }
803 } 758 }
804 return r; 759 return r;
805 } 760 }
806 761
807 /** 762 /**
808 * htb_lookup_leaf - returns next leaf class i 763 * htb_lookup_leaf - returns next leaf class in DRR order
809 * @hprio: the current one <<
810 * @prio: which prio in class <<
811 * 764 *
812 * Find leaf where current feed pointers point 765 * Find leaf where current feed pointers points to.
813 */ 766 */
814 static struct htb_class *htb_lookup_leaf(struc 767 static struct htb_class *htb_lookup_leaf(struct htb_prio *hprio, const int prio)
815 { 768 {
816 int i; 769 int i;
817 struct { 770 struct {
818 struct rb_node *root; 771 struct rb_node *root;
819 struct rb_node **pptr; 772 struct rb_node **pptr;
820 u32 *pid; 773 u32 *pid;
821 } stk[TC_HTB_MAXDEPTH], *sp = stk; 774 } stk[TC_HTB_MAXDEPTH], *sp = stk;
822 775
823 BUG_ON(!hprio->row.rb_node); 776 BUG_ON(!hprio->row.rb_node);
824 sp->root = hprio->row.rb_node; 777 sp->root = hprio->row.rb_node;
825 sp->pptr = &hprio->ptr; 778 sp->pptr = &hprio->ptr;
826 sp->pid = &hprio->last_ptr_id; 779 sp->pid = &hprio->last_ptr_id;
827 780
828 for (i = 0; i < 65535; i++) { 781 for (i = 0; i < 65535; i++) {
829 if (!*sp->pptr && *sp->pid) { 782 if (!*sp->pptr && *sp->pid) {
830 /* ptr was invalidated 783 /* ptr was invalidated but id is valid - try to recover
831 * the original or nex 784 * the original or next ptr
832 */ 785 */
833 *sp->pptr = 786 *sp->pptr =
834 htb_id_find_next_u 787 htb_id_find_next_upper(prio, sp->root, *sp->pid);
835 } 788 }
836 *sp->pid = 0; /* ptr is vali 789 *sp->pid = 0; /* ptr is valid now so that remove this hint as it
837 * can become 790 * can become out of date quickly
838 */ 791 */
839 if (!*sp->pptr) { /* we 792 if (!*sp->pptr) { /* we are at right end; rewind & go up */
840 *sp->pptr = sp->root; 793 *sp->pptr = sp->root;
841 while ((*sp->pptr)->rb 794 while ((*sp->pptr)->rb_left)
842 *sp->pptr = (* 795 *sp->pptr = (*sp->pptr)->rb_left;
843 if (sp > stk) { 796 if (sp > stk) {
844 sp--; 797 sp--;
845 if (!*sp->pptr 798 if (!*sp->pptr) {
846 WARN_O 799 WARN_ON(1);
847 return 800 return NULL;
848 } 801 }
849 htb_next_rb_no 802 htb_next_rb_node(sp->pptr);
850 } 803 }
851 } else { 804 } else {
852 struct htb_class *cl; 805 struct htb_class *cl;
853 struct htb_prio *clp; 806 struct htb_prio *clp;
854 807
855 cl = rb_entry(*sp->ppt 808 cl = rb_entry(*sp->pptr, struct htb_class, node[prio]);
856 if (!cl->level) 809 if (!cl->level)
857 return cl; 810 return cl;
858 clp = &cl->inner.clpri 811 clp = &cl->inner.clprio[prio];
859 (++sp)->root = clp->fe 812 (++sp)->root = clp->feed.rb_node;
860 sp->pptr = &clp->ptr; 813 sp->pptr = &clp->ptr;
861 sp->pid = &clp->last_p 814 sp->pid = &clp->last_ptr_id;
862 } 815 }
863 } 816 }
864 WARN_ON(1); 817 WARN_ON(1);
865 return NULL; 818 return NULL;
866 } 819 }
867 820
868 /* dequeues packet at given priority and level 821 /* dequeues packet at given priority and level; call only if
869 * you are sure that there is active class at 822 * you are sure that there is active class at prio/level
870 */ 823 */
871 static struct sk_buff *htb_dequeue_tree(struct 824 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, const int prio,
872 const 825 const int level)
873 { 826 {
874 struct sk_buff *skb = NULL; 827 struct sk_buff *skb = NULL;
875 struct htb_class *cl, *start; 828 struct htb_class *cl, *start;
876 struct htb_level *hlevel = &q->hlevel[ 829 struct htb_level *hlevel = &q->hlevel[level];
877 struct htb_prio *hprio = &hlevel->hpri 830 struct htb_prio *hprio = &hlevel->hprio[prio];
878 831
879 /* look initial class up in the row */ 832 /* look initial class up in the row */
880 start = cl = htb_lookup_leaf(hprio, pr 833 start = cl = htb_lookup_leaf(hprio, prio);
881 834
882 do { 835 do {
883 next: 836 next:
884 if (unlikely(!cl)) 837 if (unlikely(!cl))
885 return NULL; 838 return NULL;
886 839
887 /* class can be empty - it is 840 /* class can be empty - it is unlikely but can be true if leaf
888 * qdisc drops packets in enqu 841 * qdisc drops packets in enqueue routine or if someone used
889 * graft operation on the leaf 842 * graft operation on the leaf since last dequeue;
890 * simply deactivate and skip 843 * simply deactivate and skip such class
891 */ 844 */
892 if (unlikely(cl->leaf.q->q.qle 845 if (unlikely(cl->leaf.q->q.qlen == 0)) {
893 struct htb_class *next 846 struct htb_class *next;
894 htb_deactivate(q, cl); 847 htb_deactivate(q, cl);
895 848
896 /* row/level might bec 849 /* row/level might become empty */
897 if ((q->row_mask[level 850 if ((q->row_mask[level] & (1 << prio)) == 0)
898 return NULL; 851 return NULL;
899 852
900 next = htb_lookup_leaf 853 next = htb_lookup_leaf(hprio, prio);
901 854
902 if (cl == start) 855 if (cl == start) /* fix start if we just deleted it */
903 start = next; 856 start = next;
904 cl = next; 857 cl = next;
905 goto next; 858 goto next;
906 } 859 }
907 860
908 skb = cl->leaf.q->dequeue(cl-> 861 skb = cl->leaf.q->dequeue(cl->leaf.q);
909 if (likely(skb != NULL)) 862 if (likely(skb != NULL))
910 break; 863 break;
911 864
912 qdisc_warn_nonwc("htb", cl->le 865 qdisc_warn_nonwc("htb", cl->leaf.q);
913 htb_next_rb_node(level ? &cl-> 866 htb_next_rb_node(level ? &cl->parent->inner.clprio[prio].ptr:
914 &q->h 867 &q->hlevel[0].hprio[prio].ptr);
915 cl = htb_lookup_leaf(hprio, pr 868 cl = htb_lookup_leaf(hprio, prio);
916 869
917 } while (cl != start); 870 } while (cl != start);
918 871
919 if (likely(skb != NULL)) { 872 if (likely(skb != NULL)) {
920 bstats_update(&cl->bstats, skb 873 bstats_update(&cl->bstats, skb);
921 cl->leaf.deficit[level] -= qdi 874 cl->leaf.deficit[level] -= qdisc_pkt_len(skb);
922 if (cl->leaf.deficit[level] < 875 if (cl->leaf.deficit[level] < 0) {
923 cl->leaf.deficit[level 876 cl->leaf.deficit[level] += cl->quantum;
924 htb_next_rb_node(level 877 htb_next_rb_node(level ? &cl->parent->inner.clprio[prio].ptr :
925 878 &q->hlevel[0].hprio[prio].ptr);
926 } 879 }
927 /* this used to be after charg 880 /* this used to be after charge_class but this constelation
928 * gives us slightly better pe 881 * gives us slightly better performance
929 */ 882 */
930 if (!cl->leaf.q->q.qlen) 883 if (!cl->leaf.q->q.qlen)
931 htb_deactivate(q, cl); 884 htb_deactivate(q, cl);
932 htb_charge_class(q, cl, level, 885 htb_charge_class(q, cl, level, skb);
933 } 886 }
934 return skb; 887 return skb;
935 } 888 }
936 889
937 static struct sk_buff *htb_dequeue(struct Qdis 890 static struct sk_buff *htb_dequeue(struct Qdisc *sch)
938 { 891 {
939 struct sk_buff *skb; 892 struct sk_buff *skb;
940 struct htb_sched *q = qdisc_priv(sch); 893 struct htb_sched *q = qdisc_priv(sch);
941 int level; 894 int level;
942 s64 next_event; 895 s64 next_event;
943 unsigned long start_at; 896 unsigned long start_at;
944 897
945 /* try to dequeue direct packets as hi 898 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
946 skb = __qdisc_dequeue_head(&q->direct_ 899 skb = __qdisc_dequeue_head(&q->direct_queue);
947 if (skb != NULL) { 900 if (skb != NULL) {
948 ok: 901 ok:
949 qdisc_bstats_update(sch, skb); 902 qdisc_bstats_update(sch, skb);
950 qdisc_qstats_backlog_dec(sch, 903 qdisc_qstats_backlog_dec(sch, skb);
951 sch->q.qlen--; 904 sch->q.qlen--;
952 return skb; 905 return skb;
953 } 906 }
954 907
955 if (!sch->q.qlen) 908 if (!sch->q.qlen)
956 goto fin; 909 goto fin;
957 q->now = ktime_get_ns(); 910 q->now = ktime_get_ns();
958 start_at = jiffies; 911 start_at = jiffies;
959 912
960 next_event = q->now + 5LLU * NSEC_PER_ 913 next_event = q->now + 5LLU * NSEC_PER_SEC;
961 914
962 for (level = 0; level < TC_HTB_MAXDEPT 915 for (level = 0; level < TC_HTB_MAXDEPTH; level++) {
963 /* common case optimization - 916 /* common case optimization - skip event handler quickly */
964 int m; 917 int m;
965 s64 event = q->near_ev_cache[l 918 s64 event = q->near_ev_cache[level];
966 919
967 if (q->now >= event) { 920 if (q->now >= event) {
968 event = htb_do_events( 921 event = htb_do_events(q, level, start_at);
969 if (!event) 922 if (!event)
970 event = q->now 923 event = q->now + NSEC_PER_SEC;
971 q->near_ev_cache[level 924 q->near_ev_cache[level] = event;
972 } 925 }
973 926
974 if (next_event > event) 927 if (next_event > event)
975 next_event = event; 928 next_event = event;
976 929
977 m = ~q->row_mask[level]; 930 m = ~q->row_mask[level];
978 while (m != (int)(-1)) { 931 while (m != (int)(-1)) {
979 int prio = ffz(m); 932 int prio = ffz(m);
980 933
981 m |= 1 << prio; 934 m |= 1 << prio;
982 skb = htb_dequeue_tree 935 skb = htb_dequeue_tree(q, prio, level);
983 if (likely(skb != NULL 936 if (likely(skb != NULL))
984 goto ok; 937 goto ok;
985 } 938 }
986 } 939 }
>> 940 qdisc_qstats_overlimit(sch);
987 if (likely(next_event > q->now)) 941 if (likely(next_event > q->now))
988 qdisc_watchdog_schedule_ns(&q- 942 qdisc_watchdog_schedule_ns(&q->watchdog, next_event);
989 else 943 else
990 schedule_work(&q->work); 944 schedule_work(&q->work);
991 fin: 945 fin:
992 return skb; 946 return skb;
993 } 947 }
994 948
995 /* reset all classes */ 949 /* reset all classes */
996 /* always caled under BH & queue lock */ 950 /* always caled under BH & queue lock */
997 static void htb_reset(struct Qdisc *sch) 951 static void htb_reset(struct Qdisc *sch)
998 { 952 {
999 struct htb_sched *q = qdisc_priv(sch); 953 struct htb_sched *q = qdisc_priv(sch);
1000 struct htb_class *cl; 954 struct htb_class *cl;
1001 unsigned int i; 955 unsigned int i;
1002 956
1003 for (i = 0; i < q->clhash.hashsize; i 957 for (i = 0; i < q->clhash.hashsize; i++) {
1004 hlist_for_each_entry(cl, &q-> 958 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
1005 if (cl->level) 959 if (cl->level)
1006 memset(&cl->i 960 memset(&cl->inner, 0, sizeof(cl->inner));
1007 else { 961 else {
1008 if (cl->leaf. !! 962 if (cl->leaf.q)
1009 qdisc 963 qdisc_reset(cl->leaf.q);
1010 } 964 }
1011 cl->prio_activity = 0 965 cl->prio_activity = 0;
1012 cl->cmode = HTB_CAN_S 966 cl->cmode = HTB_CAN_SEND;
1013 } 967 }
1014 } 968 }
1015 qdisc_watchdog_cancel(&q->watchdog); 969 qdisc_watchdog_cancel(&q->watchdog);
1016 __qdisc_reset_queue(&q->direct_queue) 970 __qdisc_reset_queue(&q->direct_queue);
>> 971 sch->q.qlen = 0;
>> 972 sch->qstats.backlog = 0;
1017 memset(q->hlevel, 0, sizeof(q->hlevel 973 memset(q->hlevel, 0, sizeof(q->hlevel));
1018 memset(q->row_mask, 0, sizeof(q->row_ 974 memset(q->row_mask, 0, sizeof(q->row_mask));
1019 } 975 }
1020 976
1021 static const struct nla_policy htb_policy[TCA 977 static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = {
1022 [TCA_HTB_PARMS] = { .len = sizeof(str 978 [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) },
1023 [TCA_HTB_INIT] = { .len = sizeof(str 979 [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) },
1024 [TCA_HTB_CTAB] = { .type = NLA_BINAR 980 [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1025 [TCA_HTB_RTAB] = { .type = NLA_BINAR 981 [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1026 [TCA_HTB_DIRECT_QLEN] = { .type = NLA 982 [TCA_HTB_DIRECT_QLEN] = { .type = NLA_U32 },
1027 [TCA_HTB_RATE64] = { .type = NLA_U64 983 [TCA_HTB_RATE64] = { .type = NLA_U64 },
1028 [TCA_HTB_CEIL64] = { .type = NLA_U64 984 [TCA_HTB_CEIL64] = { .type = NLA_U64 },
1029 [TCA_HTB_OFFLOAD] = { .type = NLA_FLA <<
1030 }; 985 };
1031 986
1032 static void htb_work_func(struct work_struct 987 static void htb_work_func(struct work_struct *work)
1033 { 988 {
1034 struct htb_sched *q = container_of(wo 989 struct htb_sched *q = container_of(work, struct htb_sched, work);
1035 struct Qdisc *sch = q->watchdog.qdisc 990 struct Qdisc *sch = q->watchdog.qdisc;
1036 991
1037 rcu_read_lock(); 992 rcu_read_lock();
1038 __netif_schedule(qdisc_root(sch)); 993 __netif_schedule(qdisc_root(sch));
1039 rcu_read_unlock(); 994 rcu_read_unlock();
1040 } 995 }
1041 996
1042 static int htb_offload(struct net_device *dev <<
1043 { <<
1044 return dev->netdev_ops->ndo_setup_tc( <<
1045 } <<
1046 <<
1047 static int htb_init(struct Qdisc *sch, struct 997 static int htb_init(struct Qdisc *sch, struct nlattr *opt,
1048 struct netlink_ext_ack *e 998 struct netlink_ext_ack *extack)
1049 { 999 {
1050 struct net_device *dev = qdisc_dev(sc <<
1051 struct tc_htb_qopt_offload offload_op <<
1052 struct htb_sched *q = qdisc_priv(sch) 1000 struct htb_sched *q = qdisc_priv(sch);
1053 struct nlattr *tb[TCA_HTB_MAX + 1]; 1001 struct nlattr *tb[TCA_HTB_MAX + 1];
1054 struct tc_htb_glob *gopt; 1002 struct tc_htb_glob *gopt;
1055 unsigned int ntx; <<
1056 bool offload; <<
1057 int err; 1003 int err;
1058 1004
1059 qdisc_watchdog_init(&q->watchdog, sch 1005 qdisc_watchdog_init(&q->watchdog, sch);
1060 INIT_WORK(&q->work, htb_work_func); 1006 INIT_WORK(&q->work, htb_work_func);
1061 1007
1062 if (!opt) 1008 if (!opt)
1063 return -EINVAL; 1009 return -EINVAL;
1064 1010
1065 err = tcf_block_get(&q->block, &q->fi 1011 err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
1066 if (err) 1012 if (err)
1067 return err; 1013 return err;
1068 1014
1069 err = nla_parse_nested_deprecated(tb, !! 1015 err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy, NULL);
1070 NUL <<
1071 if (err < 0) 1016 if (err < 0)
1072 return err; 1017 return err;
1073 1018
1074 if (!tb[TCA_HTB_INIT]) 1019 if (!tb[TCA_HTB_INIT])
1075 return -EINVAL; 1020 return -EINVAL;
1076 1021
1077 gopt = nla_data(tb[TCA_HTB_INIT]); 1022 gopt = nla_data(tb[TCA_HTB_INIT]);
1078 if (gopt->version != HTB_VER >> 16) 1023 if (gopt->version != HTB_VER >> 16)
1079 return -EINVAL; 1024 return -EINVAL;
1080 1025
1081 offload = nla_get_flag(tb[TCA_HTB_OFF <<
1082 <<
1083 if (offload) { <<
1084 if (sch->parent != TC_H_ROOT) <<
1085 NL_SET_ERR_MSG(extack <<
1086 return -EOPNOTSUPP; <<
1087 } <<
1088 <<
1089 if (!tc_can_offload(dev) || ! <<
1090 NL_SET_ERR_MSG(extack <<
1091 return -EOPNOTSUPP; <<
1092 } <<
1093 <<
1094 q->num_direct_qdiscs = dev->r <<
1095 q->direct_qdiscs = kcalloc(q- <<
1096 si <<
1097 GF <<
1098 if (!q->direct_qdiscs) <<
1099 return -ENOMEM; <<
1100 } <<
1101 <<
1102 err = qdisc_class_hash_init(&q->clhas 1026 err = qdisc_class_hash_init(&q->clhash);
1103 if (err < 0) 1027 if (err < 0)
1104 return err; 1028 return err;
1105 1029
>> 1030 qdisc_skb_head_init(&q->direct_queue);
>> 1031
1106 if (tb[TCA_HTB_DIRECT_QLEN]) 1032 if (tb[TCA_HTB_DIRECT_QLEN])
1107 q->direct_qlen = nla_get_u32( 1033 q->direct_qlen = nla_get_u32(tb[TCA_HTB_DIRECT_QLEN]);
1108 else 1034 else
1109 q->direct_qlen = qdisc_dev(sc 1035 q->direct_qlen = qdisc_dev(sch)->tx_queue_len;
1110 1036
1111 if ((q->rate2quantum = gopt->rate2qua 1037 if ((q->rate2quantum = gopt->rate2quantum) < 1)
1112 q->rate2quantum = 1; 1038 q->rate2quantum = 1;
1113 q->defcls = gopt->defcls; 1039 q->defcls = gopt->defcls;
1114 1040
1115 if (!offload) <<
1116 return 0; <<
1117 <<
1118 for (ntx = 0; ntx < q->num_direct_qdi <<
1119 struct netdev_queue *dev_queu <<
1120 struct Qdisc *qdisc; <<
1121 <<
1122 qdisc = qdisc_create_dflt(dev <<
1123 TC_ <<
1124 if (!qdisc) { <<
1125 return -ENOMEM; <<
1126 } <<
1127 <<
1128 q->direct_qdiscs[ntx] = qdisc <<
1129 qdisc->flags |= TCQ_F_ONETXQU <<
1130 } <<
1131 <<
1132 sch->flags |= TCQ_F_MQROOT; <<
1133 <<
1134 offload_opt = (struct tc_htb_qopt_off <<
1135 .command = TC_HTB_CREATE, <<
1136 .parent_classid = TC_H_MAJ(sc <<
1137 .classid = TC_H_MIN(q->defcls <<
1138 .extack = extack, <<
1139 }; <<
1140 err = htb_offload(dev, &offload_opt); <<
1141 if (err) <<
1142 return err; <<
1143 <<
1144 /* Defer this assignment, so that htb <<
1145 * parts (especially calling ndo_setu <<
1146 */ <<
1147 q->offload = true; <<
1148 <<
1149 return 0; 1041 return 0;
1150 } 1042 }
1151 1043
1152 static void htb_attach_offload(struct Qdisc * <<
1153 { <<
1154 struct net_device *dev = qdisc_dev(sc <<
1155 struct htb_sched *q = qdisc_priv(sch) <<
1156 unsigned int ntx; <<
1157 <<
1158 for (ntx = 0; ntx < q->num_direct_qdi <<
1159 struct Qdisc *old, *qdisc = q <<
1160 <<
1161 old = dev_graft_qdisc(qdisc-> <<
1162 qdisc_put(old); <<
1163 qdisc_hash_add(qdisc, false); <<
1164 } <<
1165 for (ntx = q->num_direct_qdiscs; ntx <<
1166 struct netdev_queue *dev_queu <<
1167 struct Qdisc *old = dev_graft <<
1168 <<
1169 qdisc_put(old); <<
1170 } <<
1171 <<
1172 kfree(q->direct_qdiscs); <<
1173 q->direct_qdiscs = NULL; <<
1174 } <<
1175 <<
1176 static void htb_attach_software(struct Qdisc <<
1177 { <<
1178 struct net_device *dev = qdisc_dev(sc <<
1179 unsigned int ntx; <<
1180 <<
1181 /* Resemble qdisc_graft behavior. */ <<
1182 for (ntx = 0; ntx < dev->num_tx_queue <<
1183 struct netdev_queue *dev_queu <<
1184 struct Qdisc *old = dev_graft <<
1185 <<
1186 qdisc_refcount_inc(sch); <<
1187 <<
1188 qdisc_put(old); <<
1189 } <<
1190 } <<
1191 <<
1192 static void htb_attach(struct Qdisc *sch) <<
1193 { <<
1194 struct htb_sched *q = qdisc_priv(sch) <<
1195 <<
1196 if (q->offload) <<
1197 htb_attach_offload(sch); <<
1198 else <<
1199 htb_attach_software(sch); <<
1200 } <<
1201 <<
1202 static int htb_dump(struct Qdisc *sch, struct 1044 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1203 { 1045 {
1204 struct htb_sched *q = qdisc_priv(sch) 1046 struct htb_sched *q = qdisc_priv(sch);
1205 struct nlattr *nest; 1047 struct nlattr *nest;
1206 struct tc_htb_glob gopt; 1048 struct tc_htb_glob gopt;
1207 1049
1208 if (q->offload) <<
1209 sch->flags |= TCQ_F_OFFLOADED <<
1210 else <<
1211 sch->flags &= ~TCQ_F_OFFLOADE <<
1212 <<
1213 sch->qstats.overlimits = q->overlimit <<
1214 /* Its safe to not acquire qdisc lock 1050 /* Its safe to not acquire qdisc lock. As we hold RTNL,
1215 * no change can happen on the qdisc 1051 * no change can happen on the qdisc parameters.
1216 */ 1052 */
1217 1053
1218 gopt.direct_pkts = q->direct_pkts; 1054 gopt.direct_pkts = q->direct_pkts;
1219 gopt.version = HTB_VER; 1055 gopt.version = HTB_VER;
1220 gopt.rate2quantum = q->rate2quantum; 1056 gopt.rate2quantum = q->rate2quantum;
1221 gopt.defcls = q->defcls; 1057 gopt.defcls = q->defcls;
1222 gopt.debug = 0; 1058 gopt.debug = 0;
1223 1059
1224 nest = nla_nest_start_noflag(skb, TCA !! 1060 nest = nla_nest_start(skb, TCA_OPTIONS);
1225 if (nest == NULL) 1061 if (nest == NULL)
1226 goto nla_put_failure; 1062 goto nla_put_failure;
1227 if (nla_put(skb, TCA_HTB_INIT, sizeof 1063 if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt) ||
1228 nla_put_u32(skb, TCA_HTB_DIRECT_Q 1064 nla_put_u32(skb, TCA_HTB_DIRECT_QLEN, q->direct_qlen))
1229 goto nla_put_failure; 1065 goto nla_put_failure;
1230 if (q->offload && nla_put_flag(skb, T <<
1231 goto nla_put_failure; <<
1232 1066
1233 return nla_nest_end(skb, nest); 1067 return nla_nest_end(skb, nest);
1234 1068
1235 nla_put_failure: 1069 nla_put_failure:
1236 nla_nest_cancel(skb, nest); 1070 nla_nest_cancel(skb, nest);
1237 return -1; 1071 return -1;
1238 } 1072 }
1239 1073
1240 static int htb_dump_class(struct Qdisc *sch, 1074 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1241 struct sk_buff *skb 1075 struct sk_buff *skb, struct tcmsg *tcm)
1242 { 1076 {
1243 struct htb_class *cl = (struct htb_cl 1077 struct htb_class *cl = (struct htb_class *)arg;
1244 struct htb_sched *q = qdisc_priv(sch) <<
1245 struct nlattr *nest; 1078 struct nlattr *nest;
1246 struct tc_htb_opt opt; 1079 struct tc_htb_opt opt;
1247 1080
1248 /* Its safe to not acquire qdisc lock 1081 /* Its safe to not acquire qdisc lock. As we hold RTNL,
1249 * no change can happen on the class 1082 * no change can happen on the class parameters.
1250 */ 1083 */
1251 tcm->tcm_parent = cl->parent ? cl->pa 1084 tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT;
1252 tcm->tcm_handle = cl->common.classid; 1085 tcm->tcm_handle = cl->common.classid;
1253 if (!cl->level && cl->leaf.q) 1086 if (!cl->level && cl->leaf.q)
1254 tcm->tcm_info = cl->leaf.q->h 1087 tcm->tcm_info = cl->leaf.q->handle;
1255 1088
1256 nest = nla_nest_start_noflag(skb, TCA !! 1089 nest = nla_nest_start(skb, TCA_OPTIONS);
1257 if (nest == NULL) 1090 if (nest == NULL)
1258 goto nla_put_failure; 1091 goto nla_put_failure;
1259 1092
1260 memset(&opt, 0, sizeof(opt)); 1093 memset(&opt, 0, sizeof(opt));
1261 1094
1262 psched_ratecfg_getrate(&opt.rate, &cl 1095 psched_ratecfg_getrate(&opt.rate, &cl->rate);
1263 opt.buffer = PSCHED_NS2TICKS(cl->buff 1096 opt.buffer = PSCHED_NS2TICKS(cl->buffer);
1264 psched_ratecfg_getrate(&opt.ceil, &cl 1097 psched_ratecfg_getrate(&opt.ceil, &cl->ceil);
1265 opt.cbuffer = PSCHED_NS2TICKS(cl->cbu 1098 opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer);
1266 opt.quantum = cl->quantum; 1099 opt.quantum = cl->quantum;
1267 opt.prio = cl->prio; 1100 opt.prio = cl->prio;
1268 opt.level = cl->level; 1101 opt.level = cl->level;
1269 if (nla_put(skb, TCA_HTB_PARMS, sizeo 1102 if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt))
1270 goto nla_put_failure; 1103 goto nla_put_failure;
1271 if (q->offload && nla_put_flag(skb, T <<
1272 goto nla_put_failure; <<
1273 if ((cl->rate.rate_bytes_ps >= (1ULL 1104 if ((cl->rate.rate_bytes_ps >= (1ULL << 32)) &&
1274 nla_put_u64_64bit(skb, TCA_HTB_RA 1105 nla_put_u64_64bit(skb, TCA_HTB_RATE64, cl->rate.rate_bytes_ps,
1275 TCA_HTB_PAD)) 1106 TCA_HTB_PAD))
1276 goto nla_put_failure; 1107 goto nla_put_failure;
1277 if ((cl->ceil.rate_bytes_ps >= (1ULL 1108 if ((cl->ceil.rate_bytes_ps >= (1ULL << 32)) &&
1278 nla_put_u64_64bit(skb, TCA_HTB_CE 1109 nla_put_u64_64bit(skb, TCA_HTB_CEIL64, cl->ceil.rate_bytes_ps,
1279 TCA_HTB_PAD)) 1110 TCA_HTB_PAD))
1280 goto nla_put_failure; 1111 goto nla_put_failure;
1281 1112
1282 return nla_nest_end(skb, nest); 1113 return nla_nest_end(skb, nest);
1283 1114
1284 nla_put_failure: 1115 nla_put_failure:
1285 nla_nest_cancel(skb, nest); 1116 nla_nest_cancel(skb, nest);
1286 return -1; 1117 return -1;
1287 } 1118 }
1288 1119
1289 static void htb_offload_aggregate_stats(struc <<
1290 struc <<
1291 { <<
1292 u64 bytes = 0, packets = 0; <<
1293 struct htb_class *c; <<
1294 unsigned int i; <<
1295 <<
1296 gnet_stats_basic_sync_init(&cl->bstat <<
1297 <<
1298 for (i = 0; i < q->clhash.hashsize; i <<
1299 hlist_for_each_entry(c, &q->c <<
1300 struct htb_class *p = <<
1301 <<
1302 while (p && p->level <<
1303 p = p->parent <<
1304 <<
1305 if (p != cl) <<
1306 continue; <<
1307 <<
1308 bytes += u64_stats_re <<
1309 packets += u64_stats_ <<
1310 if (c->level == 0) { <<
1311 bytes += u64_ <<
1312 packets += u6 <<
1313 } <<
1314 } <<
1315 } <<
1316 _bstats_update(&cl->bstats, bytes, pa <<
1317 } <<
1318 <<
1319 static int 1120 static int
1320 htb_dump_class_stats(struct Qdisc *sch, unsig 1121 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d)
1321 { 1122 {
1322 struct htb_class *cl = (struct htb_cl 1123 struct htb_class *cl = (struct htb_class *)arg;
1323 struct htb_sched *q = qdisc_priv(sch) <<
1324 struct gnet_stats_queue qs = { 1124 struct gnet_stats_queue qs = {
1325 .drops = cl->drops, 1125 .drops = cl->drops,
1326 .overlimits = cl->overlimits, 1126 .overlimits = cl->overlimits,
1327 }; 1127 };
1328 __u32 qlen = 0; 1128 __u32 qlen = 0;
1329 1129
1330 if (!cl->level && cl->leaf.q) !! 1130 if (!cl->level && cl->leaf.q) {
1331 qdisc_qstats_qlen_backlog(cl- !! 1131 qlen = cl->leaf.q->q.qlen;
1332 !! 1132 qs.backlog = cl->leaf.q->qstats.backlog;
>> 1133 }
1333 cl->xstats.tokens = clamp_t(s64, PSCH 1134 cl->xstats.tokens = clamp_t(s64, PSCHED_NS2TICKS(cl->tokens),
1334 INT_MIN, 1135 INT_MIN, INT_MAX);
1335 cl->xstats.ctokens = clamp_t(s64, PSC 1136 cl->xstats.ctokens = clamp_t(s64, PSCHED_NS2TICKS(cl->ctokens),
1336 INT_MIN, 1137 INT_MIN, INT_MAX);
1337 1138
1338 if (q->offload) { !! 1139 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
1339 if (!cl->level) { !! 1140 d, NULL, &cl->bstats) < 0 ||
1340 if (cl->leaf.q) <<
1341 cl->bstats = <<
1342 else <<
1343 gnet_stats_ba <<
1344 _bstats_update(&cl->b <<
1345 u64_st <<
1346 u64_st <<
1347 } else { <<
1348 htb_offload_aggregate <<
1349 } <<
1350 } <<
1351 <<
1352 if (gnet_stats_copy_basic(d, NULL, &c <<
1353 gnet_stats_copy_rate_est(d, &cl-> 1141 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1354 gnet_stats_copy_queue(d, NULL, &q 1142 gnet_stats_copy_queue(d, NULL, &qs, qlen) < 0)
1355 return -1; 1143 return -1;
1356 1144
1357 return gnet_stats_copy_app(d, &cl->xs 1145 return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1358 } 1146 }
1359 1147
1360 static struct netdev_queue * <<
1361 htb_select_queue(struct Qdisc *sch, struct tc <<
1362 { <<
1363 struct net_device *dev = qdisc_dev(sc <<
1364 struct tc_htb_qopt_offload offload_op <<
1365 struct htb_sched *q = qdisc_priv(sch) <<
1366 int err; <<
1367 <<
1368 if (!q->offload) <<
1369 return sch->dev_queue; <<
1370 <<
1371 offload_opt = (struct tc_htb_qopt_off <<
1372 .command = TC_HTB_LEAF_QUERY_ <<
1373 .classid = TC_H_MIN(tcm->tcm_ <<
1374 }; <<
1375 err = htb_offload(dev, &offload_opt); <<
1376 if (err || offload_opt.qid >= dev->nu <<
1377 return NULL; <<
1378 return netdev_get_tx_queue(dev, offlo <<
1379 } <<
1380 <<
1381 static struct Qdisc * <<
1382 htb_graft_helper(struct netdev_queue *dev_que <<
1383 { <<
1384 struct net_device *dev = dev_queue->d <<
1385 struct Qdisc *old_q; <<
1386 <<
1387 if (dev->flags & IFF_UP) <<
1388 dev_deactivate(dev); <<
1389 old_q = dev_graft_qdisc(dev_queue, ne <<
1390 if (new_q) <<
1391 new_q->flags |= TCQ_F_ONETXQU <<
1392 if (dev->flags & IFF_UP) <<
1393 dev_activate(dev); <<
1394 <<
1395 return old_q; <<
1396 } <<
1397 <<
1398 static struct netdev_queue *htb_offload_get_q <<
1399 { <<
1400 struct netdev_queue *queue; <<
1401 <<
1402 queue = cl->leaf.offload_queue; <<
1403 if (!(cl->leaf.q->flags & TCQ_F_BUILT <<
1404 WARN_ON(cl->leaf.q->dev_queue <<
1405 <<
1406 return queue; <<
1407 } <<
1408 <<
1409 static void htb_offload_move_qdisc(struct Qdi <<
1410 struct htb <<
1411 { <<
1412 struct netdev_queue *queue_old, *queu <<
1413 struct net_device *dev = qdisc_dev(sc <<
1414 <<
1415 queue_old = htb_offload_get_queue(cl_ <<
1416 queue_new = htb_offload_get_queue(cl_ <<
1417 <<
1418 if (!destroying) { <<
1419 struct Qdisc *qdisc; <<
1420 <<
1421 if (dev->flags & IFF_UP) <<
1422 dev_deactivate(dev); <<
1423 qdisc = dev_graft_qdisc(queue <<
1424 WARN_ON(qdisc != cl_old->leaf <<
1425 } <<
1426 <<
1427 if (!(cl_old->leaf.q->flags & TCQ_F_B <<
1428 cl_old->leaf.q->dev_queue = q <<
1429 cl_old->leaf.offload_queue = queue_ne <<
1430 <<
1431 if (!destroying) { <<
1432 struct Qdisc *qdisc; <<
1433 <<
1434 qdisc = dev_graft_qdisc(queue <<
1435 if (dev->flags & IFF_UP) <<
1436 dev_activate(dev); <<
1437 WARN_ON(!(qdisc->flags & TCQ_ <<
1438 } <<
1439 } <<
1440 <<
1441 static int htb_graft(struct Qdisc *sch, unsig 1148 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1442 struct Qdisc **old, stru 1149 struct Qdisc **old, struct netlink_ext_ack *extack)
1443 { 1150 {
1444 struct netdev_queue *dev_queue = sch- <<
1445 struct htb_class *cl = (struct htb_cl 1151 struct htb_class *cl = (struct htb_class *)arg;
1446 struct htb_sched *q = qdisc_priv(sch) <<
1447 struct Qdisc *old_q; <<
1448 1152
1449 if (cl->level) 1153 if (cl->level)
1450 return -EINVAL; 1154 return -EINVAL;
1451 !! 1155 if (new == NULL &&
1452 if (q->offload) !! 1156 (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
1453 dev_queue = htb_offload_get_q !! 1157 cl->common.classid, extack)) == NULL)
1454 !! 1158 return -ENOBUFS;
1455 if (!new) { <<
1456 new = qdisc_create_dflt(dev_q <<
1457 cl->c <<
1458 if (!new) <<
1459 return -ENOBUFS; <<
1460 } <<
1461 <<
1462 if (q->offload) { <<
1463 /* One ref for cl->leaf.q, th <<
1464 qdisc_refcount_inc(new); <<
1465 old_q = htb_graft_helper(dev_ <<
1466 } <<
1467 1159
1468 *old = qdisc_replace(sch, new, &cl->l 1160 *old = qdisc_replace(sch, new, &cl->leaf.q);
1469 <<
1470 if (q->offload) { <<
1471 WARN_ON(old_q != *old); <<
1472 qdisc_put(old_q); <<
1473 } <<
1474 <<
1475 return 0; 1161 return 0;
1476 } 1162 }
1477 1163
1478 static struct Qdisc *htb_leaf(struct Qdisc *s 1164 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg)
1479 { 1165 {
1480 struct htb_class *cl = (struct htb_cl 1166 struct htb_class *cl = (struct htb_class *)arg;
1481 return !cl->level ? cl->leaf.q : NULL 1167 return !cl->level ? cl->leaf.q : NULL;
1482 } 1168 }
1483 1169
1484 static void htb_qlen_notify(struct Qdisc *sch 1170 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg)
1485 { 1171 {
1486 struct htb_class *cl = (struct htb_cl 1172 struct htb_class *cl = (struct htb_class *)arg;
1487 1173
1488 htb_deactivate(qdisc_priv(sch), cl); 1174 htb_deactivate(qdisc_priv(sch), cl);
1489 } 1175 }
1490 1176
1491 static inline int htb_parent_last_child(struc 1177 static inline int htb_parent_last_child(struct htb_class *cl)
1492 { 1178 {
1493 if (!cl->parent) 1179 if (!cl->parent)
1494 /* the root class */ 1180 /* the root class */
1495 return 0; 1181 return 0;
1496 if (cl->parent->children > 1) 1182 if (cl->parent->children > 1)
1497 /* not the last child */ 1183 /* not the last child */
1498 return 0; 1184 return 0;
1499 return 1; 1185 return 1;
1500 } 1186 }
1501 1187
1502 static void htb_parent_to_leaf(struct Qdisc * !! 1188 static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl,
1503 struct Qdisc * 1189 struct Qdisc *new_q)
1504 { 1190 {
1505 struct htb_sched *q = qdisc_priv(sch) <<
1506 struct htb_class *parent = cl->parent 1191 struct htb_class *parent = cl->parent;
1507 1192
1508 WARN_ON(cl->level || !cl->leaf.q || c 1193 WARN_ON(cl->level || !cl->leaf.q || cl->prio_activity);
1509 1194
1510 if (parent->cmode != HTB_CAN_SEND) 1195 if (parent->cmode != HTB_CAN_SEND)
1511 htb_safe_rb_erase(&parent->pq 1196 htb_safe_rb_erase(&parent->pq_node,
1512 &q->hlevel[ 1197 &q->hlevel[parent->level].wait_pq);
1513 1198
1514 parent->level = 0; 1199 parent->level = 0;
1515 memset(&parent->inner, 0, sizeof(pare 1200 memset(&parent->inner, 0, sizeof(parent->inner));
1516 parent->leaf.q = new_q ? new_q : &noo 1201 parent->leaf.q = new_q ? new_q : &noop_qdisc;
1517 parent->tokens = parent->buffer; 1202 parent->tokens = parent->buffer;
1518 parent->ctokens = parent->cbuffer; 1203 parent->ctokens = parent->cbuffer;
1519 parent->t_c = ktime_get_ns(); 1204 parent->t_c = ktime_get_ns();
1520 parent->cmode = HTB_CAN_SEND; 1205 parent->cmode = HTB_CAN_SEND;
1521 if (q->offload) <<
1522 parent->leaf.offload_queue = <<
1523 } <<
1524 <<
1525 static void htb_parent_to_leaf_offload(struct <<
1526 struct <<
1527 struct <<
1528 { <<
1529 struct Qdisc *old_q; <<
1530 <<
1531 /* One ref for cl->leaf.q, the other <<
1532 if (new_q) <<
1533 qdisc_refcount_inc(new_q); <<
1534 old_q = htb_graft_helper(dev_queue, n <<
1535 WARN_ON(!(old_q->flags & TCQ_F_BUILTI <<
1536 } <<
1537 <<
1538 static int htb_destroy_class_offload(struct Q <<
1539 bool las <<
1540 struct n <<
1541 { <<
1542 struct tc_htb_qopt_offload offload_op <<
1543 struct netdev_queue *dev_queue; <<
1544 struct Qdisc *q = cl->leaf.q; <<
1545 struct Qdisc *old; <<
1546 int err; <<
1547 <<
1548 if (cl->level) <<
1549 return -EINVAL; <<
1550 <<
1551 WARN_ON(!q); <<
1552 dev_queue = htb_offload_get_queue(cl) <<
1553 /* When destroying, caller qdisc_graf <<
1554 * qdisc_put for the qdisc being dest <<
1555 * does not need to graft or qdisc_pu <<
1556 */ <<
1557 if (!destroying) { <<
1558 old = htb_graft_helper(dev_qu <<
1559 /* Last qdisc grafted should <<
1560 * calling htb_delete. <<
1561 */ <<
1562 WARN_ON(old != q); <<
1563 } <<
1564 <<
1565 if (cl->parent) { <<
1566 _bstats_update(&cl->parent->b <<
1567 u64_stats_read <<
1568 u64_stats_read <<
1569 } <<
1570 <<
1571 offload_opt = (struct tc_htb_qopt_off <<
1572 .command = !last_child ? TC_H <<
1573 destroying ? TC_HT <<
1574 TC_HTB_LEAF_DEL_LA <<
1575 .classid = cl->common.classid <<
1576 .extack = extack, <<
1577 }; <<
1578 err = htb_offload(qdisc_dev(sch), &of <<
1579 <<
1580 if (!destroying) { <<
1581 if (!err) <<
1582 qdisc_put(old); <<
1583 else <<
1584 htb_graft_helper(dev_ <<
1585 } <<
1586 <<
1587 if (last_child) <<
1588 return err; <<
1589 <<
1590 if (!err && offload_opt.classid != TC <<
1591 u32 classid = TC_H_MAJ(sch->h <<
1592 TC_H_MIN(offloa <<
1593 struct htb_class *moved_cl = <<
1594 <<
1595 htb_offload_move_qdisc(sch, m <<
1596 } <<
1597 <<
1598 return err; <<
1599 } 1206 }
1600 1207
1601 static void htb_destroy_class(struct Qdisc *s 1208 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
1602 { 1209 {
1603 if (!cl->level) { 1210 if (!cl->level) {
1604 WARN_ON(!cl->leaf.q); 1211 WARN_ON(!cl->leaf.q);
1605 qdisc_put(cl->leaf.q); 1212 qdisc_put(cl->leaf.q);
1606 } 1213 }
1607 gen_kill_estimator(&cl->rate_est); 1214 gen_kill_estimator(&cl->rate_est);
1608 tcf_block_put(cl->block); 1215 tcf_block_put(cl->block);
1609 kfree(cl); 1216 kfree(cl);
1610 } 1217 }
1611 1218
1612 static void htb_destroy(struct Qdisc *sch) 1219 static void htb_destroy(struct Qdisc *sch)
1613 { 1220 {
1614 struct net_device *dev = qdisc_dev(sc <<
1615 struct tc_htb_qopt_offload offload_op <<
1616 struct htb_sched *q = qdisc_priv(sch) 1221 struct htb_sched *q = qdisc_priv(sch);
1617 struct hlist_node *next; 1222 struct hlist_node *next;
1618 bool nonempty, changed; <<
1619 struct htb_class *cl; 1223 struct htb_class *cl;
1620 unsigned int i; 1224 unsigned int i;
1621 1225
1622 cancel_work_sync(&q->work); 1226 cancel_work_sync(&q->work);
1623 qdisc_watchdog_cancel(&q->watchdog); 1227 qdisc_watchdog_cancel(&q->watchdog);
1624 /* This line used to be after htb_des 1228 /* This line used to be after htb_destroy_class call below
1625 * and surprisingly it worked in 2.4. 1229 * and surprisingly it worked in 2.4. But it must precede it
1626 * because filter need its target cla 1230 * because filter need its target class alive to be able to call
1627 * unbind_filter on it (without Oops) 1231 * unbind_filter on it (without Oops).
1628 */ 1232 */
1629 tcf_block_put(q->block); 1233 tcf_block_put(q->block);
1630 1234
1631 for (i = 0; i < q->clhash.hashsize; i 1235 for (i = 0; i < q->clhash.hashsize; i++) {
1632 hlist_for_each_entry(cl, &q-> 1236 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
1633 tcf_block_put(cl->blo 1237 tcf_block_put(cl->block);
1634 cl->block = NULL; 1238 cl->block = NULL;
1635 } 1239 }
1636 } 1240 }
1637 !! 1241 for (i = 0; i < q->clhash.hashsize; i++) {
1638 do { !! 1242 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
1639 nonempty = false; !! 1243 common.hnode)
1640 changed = false; !! 1244 htb_destroy_class(sch, cl);
1641 for (i = 0; i < q->clhash.has !! 1245 }
1642 hlist_for_each_entry_ <<
1643 <<
1644 bool last_chi <<
1645 <<
1646 if (!q->offlo <<
1647 htb_d <<
1648 conti <<
1649 } <<
1650 <<
1651 nonempty = tr <<
1652 <<
1653 if (cl->level <<
1654 conti <<
1655 <<
1656 changed = tru <<
1657 <<
1658 last_child = <<
1659 htb_destroy_c <<
1660 <<
1661 qdisc_class_h <<
1662 <<
1663 if (cl->paren <<
1664 cl->p <<
1665 if (last_chil <<
1666 htb_p <<
1667 htb_destroy_c <<
1668 } <<
1669 } <<
1670 } while (changed); <<
1671 WARN_ON(nonempty); <<
1672 <<
1673 qdisc_class_hash_destroy(&q->clhash); 1246 qdisc_class_hash_destroy(&q->clhash);
1674 __qdisc_reset_queue(&q->direct_queue) 1247 __qdisc_reset_queue(&q->direct_queue);
1675 <<
1676 if (q->offload) { <<
1677 offload_opt = (struct tc_htb_ <<
1678 .command = TC_HTB_DES <<
1679 }; <<
1680 htb_offload(dev, &offload_opt <<
1681 } <<
1682 <<
1683 if (!q->direct_qdiscs) <<
1684 return; <<
1685 for (i = 0; i < q->num_direct_qdiscs <<
1686 qdisc_put(q->direct_qdiscs[i] <<
1687 kfree(q->direct_qdiscs); <<
1688 } 1248 }
1689 1249
1690 static int htb_delete(struct Qdisc *sch, unsi !! 1250 static int htb_delete(struct Qdisc *sch, unsigned long arg)
1691 struct netlink_ext_ack <<
1692 { 1251 {
1693 struct htb_sched *q = qdisc_priv(sch) 1252 struct htb_sched *q = qdisc_priv(sch);
1694 struct htb_class *cl = (struct htb_cl 1253 struct htb_class *cl = (struct htb_class *)arg;
1695 struct Qdisc *new_q = NULL; 1254 struct Qdisc *new_q = NULL;
1696 int last_child = 0; 1255 int last_child = 0;
1697 int err; <<
1698 1256
1699 /* TODO: why don't allow to delete su 1257 /* TODO: why don't allow to delete subtree ? references ? does
1700 * tc subsys guarantee us that in htb 1258 * tc subsys guarantee us that in htb_destroy it holds no class
1701 * refs so that we can remove childre 1259 * refs so that we can remove children safely there ?
1702 */ 1260 */
1703 if (cl->children || qdisc_class_in_us !! 1261 if (cl->children || cl->filter_cnt)
1704 NL_SET_ERR_MSG(extack, "HTB c <<
1705 return -EBUSY; 1262 return -EBUSY;
1706 } <<
1707 <<
1708 if (!cl->level && htb_parent_last_chi <<
1709 last_child = 1; <<
1710 <<
1711 if (q->offload) { <<
1712 err = htb_destroy_class_offlo <<
1713 <<
1714 if (err) <<
1715 return err; <<
1716 } <<
1717 <<
1718 if (last_child) { <<
1719 struct netdev_queue *dev_queu <<
1720 1263
1721 if (q->offload) !! 1264 if (!cl->level && htb_parent_last_child(cl)) {
1722 dev_queue = htb_offlo !! 1265 new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
1723 <<
1724 new_q = qdisc_create_dflt(dev <<
1725 cl- 1266 cl->parent->common.classid,
1726 NUL 1267 NULL);
1727 if (q->offload) !! 1268 last_child = 1;
1728 htb_parent_to_leaf_of <<
1729 } 1269 }
1730 1270
1731 sch_tree_lock(sch); 1271 sch_tree_lock(sch);
1732 1272
1733 if (!cl->level) !! 1273 if (!cl->level) {
1734 qdisc_purge_queue(cl->leaf.q) !! 1274 unsigned int qlen = cl->leaf.q->q.qlen;
>> 1275 unsigned int backlog = cl->leaf.q->qstats.backlog;
>> 1276
>> 1277 qdisc_reset(cl->leaf.q);
>> 1278 qdisc_tree_reduce_backlog(cl->leaf.q, qlen, backlog);
>> 1279 }
1735 1280
1736 /* delete from hash and active; remai 1281 /* delete from hash and active; remainder in destroy_class */
1737 qdisc_class_hash_remove(&q->clhash, & 1282 qdisc_class_hash_remove(&q->clhash, &cl->common);
1738 if (cl->parent) 1283 if (cl->parent)
1739 cl->parent->children--; 1284 cl->parent->children--;
1740 1285
1741 if (cl->prio_activity) 1286 if (cl->prio_activity)
1742 htb_deactivate(q, cl); 1287 htb_deactivate(q, cl);
1743 1288
1744 if (cl->cmode != HTB_CAN_SEND) 1289 if (cl->cmode != HTB_CAN_SEND)
1745 htb_safe_rb_erase(&cl->pq_nod 1290 htb_safe_rb_erase(&cl->pq_node,
1746 &q->hlevel[ 1291 &q->hlevel[cl->level].wait_pq);
1747 1292
1748 if (last_child) 1293 if (last_child)
1749 htb_parent_to_leaf(sch, cl, n !! 1294 htb_parent_to_leaf(q, cl, new_q);
1750 1295
1751 sch_tree_unlock(sch); 1296 sch_tree_unlock(sch);
1752 1297
1753 htb_destroy_class(sch, cl); 1298 htb_destroy_class(sch, cl);
1754 return 0; 1299 return 0;
1755 } 1300 }
1756 1301
1757 static int htb_change_class(struct Qdisc *sch 1302 static int htb_change_class(struct Qdisc *sch, u32 classid,
1758 u32 parentid, str 1303 u32 parentid, struct nlattr **tca,
1759 unsigned long *ar 1304 unsigned long *arg, struct netlink_ext_ack *extack)
1760 { 1305 {
1761 int err = -EINVAL; 1306 int err = -EINVAL;
1762 struct htb_sched *q = qdisc_priv(sch) 1307 struct htb_sched *q = qdisc_priv(sch);
1763 struct htb_class *cl = (struct htb_cl 1308 struct htb_class *cl = (struct htb_class *)*arg, *parent;
1764 struct tc_htb_qopt_offload offload_op <<
1765 struct nlattr *opt = tca[TCA_OPTIONS] 1309 struct nlattr *opt = tca[TCA_OPTIONS];
1766 struct nlattr *tb[TCA_HTB_MAX + 1]; 1310 struct nlattr *tb[TCA_HTB_MAX + 1];
1767 struct Qdisc *parent_qdisc = NULL; <<
1768 struct netdev_queue *dev_queue; <<
1769 struct tc_htb_opt *hopt; 1311 struct tc_htb_opt *hopt;
1770 u64 rate64, ceil64; 1312 u64 rate64, ceil64;
1771 int warn = 0; 1313 int warn = 0;
1772 1314
1773 /* extract all subattrs from opt attr 1315 /* extract all subattrs from opt attr */
1774 if (!opt) 1316 if (!opt)
1775 goto failure; 1317 goto failure;
1776 1318
1777 err = nla_parse_nested_deprecated(tb, !! 1319 err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy, NULL);
1778 ext <<
1779 if (err < 0) 1320 if (err < 0)
1780 goto failure; 1321 goto failure;
1781 1322
1782 err = -EINVAL; 1323 err = -EINVAL;
1783 if (tb[TCA_HTB_PARMS] == NULL) 1324 if (tb[TCA_HTB_PARMS] == NULL)
1784 goto failure; 1325 goto failure;
1785 1326
1786 parent = parentid == TC_H_ROOT ? NULL 1327 parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch);
1787 1328
1788 hopt = nla_data(tb[TCA_HTB_PARMS]); 1329 hopt = nla_data(tb[TCA_HTB_PARMS]);
1789 if (!hopt->rate.rate || !hopt->ceil.r 1330 if (!hopt->rate.rate || !hopt->ceil.rate)
1790 goto failure; 1331 goto failure;
1791 1332
1792 if (q->offload) { <<
1793 /* Options not supported by t <<
1794 if (hopt->rate.overhead || ho <<
1795 NL_SET_ERR_MSG(extack <<
1796 goto failure; <<
1797 } <<
1798 if (hopt->rate.mpu || hopt->c <<
1799 NL_SET_ERR_MSG(extack <<
1800 goto failure; <<
1801 } <<
1802 } <<
1803 <<
1804 /* Keeping backward compatible with r 1333 /* Keeping backward compatible with rate_table based iproute2 tc */
1805 if (hopt->rate.linklayer == TC_LINKLA 1334 if (hopt->rate.linklayer == TC_LINKLAYER_UNAWARE)
1806 qdisc_put_rtab(qdisc_get_rtab 1335 qdisc_put_rtab(qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB],
1807 1336 NULL));
1808 1337
1809 if (hopt->ceil.linklayer == TC_LINKLA 1338 if (hopt->ceil.linklayer == TC_LINKLAYER_UNAWARE)
1810 qdisc_put_rtab(qdisc_get_rtab 1339 qdisc_put_rtab(qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB],
1811 1340 NULL));
1812 1341
1813 rate64 = tb[TCA_HTB_RATE64] ? nla_get <<
1814 ceil64 = tb[TCA_HTB_CEIL64] ? nla_get <<
1815 <<
1816 if (!cl) { /* new class 1342 if (!cl) { /* new class */
1817 struct net_device *dev = qdis !! 1343 struct Qdisc *new_q;
1818 struct Qdisc *new_q, *old_q; <<
1819 int prio; 1344 int prio;
1820 struct { 1345 struct {
1821 struct nlattr 1346 struct nlattr nla;
1822 struct gnet_estimator 1347 struct gnet_estimator opt;
1823 } est = { 1348 } est = {
1824 .nla = { 1349 .nla = {
1825 .nla_len 1350 .nla_len = nla_attr_size(sizeof(est.opt)),
1826 .nla_type 1351 .nla_type = TCA_RATE,
1827 }, 1352 },
1828 .opt = { 1353 .opt = {
1829 /* 4s interva 1354 /* 4s interval, 16s averaging constant */
1830 .interval 1355 .interval = 2,
1831 .ewma_log 1356 .ewma_log = 2,
1832 }, 1357 },
1833 }; 1358 };
1834 1359
1835 /* check for valid classid */ 1360 /* check for valid classid */
1836 if (!classid || TC_H_MAJ(clas 1361 if (!classid || TC_H_MAJ(classid ^ sch->handle) ||
1837 htb_find(classid, sch)) 1362 htb_find(classid, sch))
1838 goto failure; 1363 goto failure;
1839 1364
1840 /* check maximal depth */ 1365 /* check maximal depth */
1841 if (parent && parent->parent 1366 if (parent && parent->parent && parent->parent->level < 2) {
1842 NL_SET_ERR_MSG_MOD(ex !! 1367 pr_err("htb: tree is too deep\n");
1843 goto failure; 1368 goto failure;
1844 } 1369 }
1845 err = -ENOBUFS; 1370 err = -ENOBUFS;
1846 cl = kzalloc(sizeof(*cl), GFP 1371 cl = kzalloc(sizeof(*cl), GFP_KERNEL);
1847 if (!cl) 1372 if (!cl)
1848 goto failure; 1373 goto failure;
1849 1374
1850 gnet_stats_basic_sync_init(&c <<
1851 gnet_stats_basic_sync_init(&c <<
1852 <<
1853 err = tcf_block_get(&cl->bloc 1375 err = tcf_block_get(&cl->block, &cl->filter_list, sch, extack);
1854 if (err) { 1376 if (err) {
1855 kfree(cl); 1377 kfree(cl);
1856 goto failure; 1378 goto failure;
1857 } 1379 }
1858 if (htb_rate_est || tca[TCA_R 1380 if (htb_rate_est || tca[TCA_RATE]) {
1859 err = gen_new_estimat 1381 err = gen_new_estimator(&cl->bstats, NULL,
1860 1382 &cl->rate_est,
1861 1383 NULL,
1862 !! 1384 qdisc_root_sleeping_running(sch),
1863 1385 tca[TCA_RATE] ? : &est.nla);
1864 if (err) !! 1386 if (err) {
1865 goto err_bloc !! 1387 tcf_block_put(cl->block);
>> 1388 kfree(cl);
>> 1389 goto failure;
>> 1390 }
1866 } 1391 }
1867 1392
1868 cl->children = 0; 1393 cl->children = 0;
1869 RB_CLEAR_NODE(&cl->pq_node); 1394 RB_CLEAR_NODE(&cl->pq_node);
1870 1395
1871 for (prio = 0; prio < TC_HTB_ 1396 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++)
1872 RB_CLEAR_NODE(&cl->no 1397 RB_CLEAR_NODE(&cl->node[prio]);
1873 1398
1874 cl->common.classid = classid; <<
1875 <<
1876 /* Make sure nothing interrup <<
1877 * ndo_setup_tc calls. <<
1878 */ <<
1879 ASSERT_RTNL(); <<
1880 <<
1881 /* create leaf qdisc early be 1399 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1882 * so that can't be used insi 1400 * so that can't be used inside of sch_tree_lock
1883 * -- thanks to Karlis Peisen 1401 * -- thanks to Karlis Peisenieks
1884 */ 1402 */
1885 if (!q->offload) { !! 1403 new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
1886 dev_queue = sch->dev_ <<
1887 } else if (!(parent && !paren <<
1888 /* Assign a dev_queue <<
1889 offload_opt = (struct <<
1890 .command = TC <<
1891 .classid = cl <<
1892 .parent_class <<
1893 TC_H_ <<
1894 TC_HT <<
1895 .rate = max_t <<
1896 .ceil = max_t <<
1897 .prio = hopt- <<
1898 .quantum = ho <<
1899 .extack = ext <<
1900 }; <<
1901 err = htb_offload(dev <<
1902 if (err) { <<
1903 NL_SET_ERR_MS <<
1904 <<
1905 goto err_kill <<
1906 } <<
1907 dev_queue = netdev_ge <<
1908 } else { /* First child. */ <<
1909 dev_queue = htb_offlo <<
1910 old_q = htb_graft_hel <<
1911 WARN_ON(old_q != pare <<
1912 offload_opt = (struct <<
1913 .command = TC <<
1914 .classid = cl <<
1915 .parent_class <<
1916 TC_H_ <<
1917 .rate = max_t <<
1918 .ceil = max_t <<
1919 .prio = hopt- <<
1920 .quantum = ho <<
1921 .extack = ext <<
1922 }; <<
1923 err = htb_offload(dev <<
1924 if (err) { <<
1925 NL_SET_ERR_MS <<
1926 <<
1927 htb_graft_hel <<
1928 goto err_kill <<
1929 } <<
1930 _bstats_update(&paren <<
1931 u64_st <<
1932 u64_st <<
1933 qdisc_put(old_q); <<
1934 } <<
1935 new_q = qdisc_create_dflt(dev <<
1936 cla 1404 classid, NULL);
1937 if (q->offload) { <<
1938 /* One ref for cl->le <<
1939 if (new_q) <<
1940 qdisc_refcoun <<
1941 old_q = htb_graft_hel <<
1942 /* No qdisc_put neede <<
1943 WARN_ON(!(old_q->flag <<
1944 } <<
1945 sch_tree_lock(sch); 1405 sch_tree_lock(sch);
1946 if (parent && !parent->level) 1406 if (parent && !parent->level) {
>> 1407 unsigned int qlen = parent->leaf.q->q.qlen;
>> 1408 unsigned int backlog = parent->leaf.q->qstats.backlog;
>> 1409
1947 /* turn parent into i 1410 /* turn parent into inner node */
1948 qdisc_purge_queue(par !! 1411 qdisc_reset(parent->leaf.q);
1949 parent_qdisc = parent !! 1412 qdisc_tree_reduce_backlog(parent->leaf.q, qlen, backlog);
>> 1413 qdisc_put(parent->leaf.q);
1950 if (parent->prio_acti 1414 if (parent->prio_activity)
1951 htb_deactivat 1415 htb_deactivate(q, parent);
1952 1416
1953 /* remove from evt li 1417 /* remove from evt list because of level change */
1954 if (parent->cmode != 1418 if (parent->cmode != HTB_CAN_SEND) {
1955 htb_safe_rb_e 1419 htb_safe_rb_erase(&parent->pq_node, &q->hlevel[0].wait_pq);
1956 parent->cmode 1420 parent->cmode = HTB_CAN_SEND;
1957 } 1421 }
1958 parent->level = (pare 1422 parent->level = (parent->parent ? parent->parent->level
1959 : TC 1423 : TC_HTB_MAXDEPTH) - 1;
1960 memset(&parent->inner 1424 memset(&parent->inner, 0, sizeof(parent->inner));
1961 } 1425 }
1962 <<
1963 /* leaf (we) needs elementary 1426 /* leaf (we) needs elementary qdisc */
1964 cl->leaf.q = new_q ? new_q : 1427 cl->leaf.q = new_q ? new_q : &noop_qdisc;
1965 if (q->offload) <<
1966 cl->leaf.offload_queu <<
1967 1428
>> 1429 cl->common.classid = classid;
1968 cl->parent = parent; 1430 cl->parent = parent;
1969 1431
1970 /* set class to be in HTB_CAN 1432 /* set class to be in HTB_CAN_SEND state */
1971 cl->tokens = PSCHED_TICKS2NS( 1433 cl->tokens = PSCHED_TICKS2NS(hopt->buffer);
1972 cl->ctokens = PSCHED_TICKS2NS 1434 cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer);
1973 cl->mbuffer = 60ULL * NSEC_PE 1435 cl->mbuffer = 60ULL * NSEC_PER_SEC; /* 1min */
1974 cl->t_c = ktime_get_ns(); 1436 cl->t_c = ktime_get_ns();
1975 cl->cmode = HTB_CAN_SEND; 1437 cl->cmode = HTB_CAN_SEND;
1976 1438
1977 /* attach to the hash list an 1439 /* attach to the hash list and parent's family */
1978 qdisc_class_hash_insert(&q->c 1440 qdisc_class_hash_insert(&q->clhash, &cl->common);
1979 if (parent) 1441 if (parent)
1980 parent->children++; 1442 parent->children++;
1981 if (cl->leaf.q != &noop_qdisc 1443 if (cl->leaf.q != &noop_qdisc)
1982 qdisc_hash_add(cl->le 1444 qdisc_hash_add(cl->leaf.q, true);
1983 } else { 1445 } else {
1984 if (tca[TCA_RATE]) { 1446 if (tca[TCA_RATE]) {
1985 err = gen_replace_est 1447 err = gen_replace_estimator(&cl->bstats, NULL,
1986 1448 &cl->rate_est,
1987 1449 NULL,
1988 !! 1450 qdisc_root_sleeping_running(sch),
1989 1451 tca[TCA_RATE]);
1990 if (err) 1452 if (err)
1991 return err; 1453 return err;
1992 } 1454 }
1993 <<
1994 if (q->offload) { <<
1995 struct net_device *de <<
1996 <<
1997 offload_opt = (struct <<
1998 .command = TC <<
1999 .classid = cl <<
2000 .rate = max_t <<
2001 .ceil = max_t <<
2002 .prio = hopt- <<
2003 .quantum = ho <<
2004 .extack = ext <<
2005 }; <<
2006 err = htb_offload(dev <<
2007 if (err) <<
2008 /* Estimator <<
2009 * as well, s <<
2010 * the estima <<
2011 * offload an <<
2012 * only when <<
2013 */ <<
2014 return err; <<
2015 } <<
2016 <<
2017 sch_tree_lock(sch); 1455 sch_tree_lock(sch);
2018 } 1456 }
2019 1457
>> 1458 rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0;
>> 1459
>> 1460 ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0;
>> 1461
2020 psched_ratecfg_precompute(&cl->rate, 1462 psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64);
2021 psched_ratecfg_precompute(&cl->ceil, 1463 psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64);
2022 1464
2023 /* it used to be a nasty bug here, we 1465 /* it used to be a nasty bug here, we have to check that node
2024 * is really leaf before changing cl- 1466 * is really leaf before changing cl->leaf !
2025 */ 1467 */
2026 if (!cl->level) { 1468 if (!cl->level) {
2027 u64 quantum = cl->rate.rate_b 1469 u64 quantum = cl->rate.rate_bytes_ps;
2028 1470
2029 do_div(quantum, q->rate2quant 1471 do_div(quantum, q->rate2quantum);
2030 cl->quantum = min_t(u64, quan 1472 cl->quantum = min_t(u64, quantum, INT_MAX);
2031 1473
2032 if (!hopt->quantum && cl->qua 1474 if (!hopt->quantum && cl->quantum < 1000) {
2033 warn = -1; 1475 warn = -1;
2034 cl->quantum = 1000; 1476 cl->quantum = 1000;
2035 } 1477 }
2036 if (!hopt->quantum && cl->qua 1478 if (!hopt->quantum && cl->quantum > 200000) {
2037 warn = 1; 1479 warn = 1;
2038 cl->quantum = 200000; 1480 cl->quantum = 200000;
2039 } 1481 }
2040 if (hopt->quantum) 1482 if (hopt->quantum)
2041 cl->quantum = hopt->q 1483 cl->quantum = hopt->quantum;
2042 if ((cl->prio = hopt->prio) > 1484 if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO)
2043 cl->prio = TC_HTB_NUM 1485 cl->prio = TC_HTB_NUMPRIO - 1;
2044 } 1486 }
2045 1487
2046 cl->buffer = PSCHED_TICKS2NS(hopt->bu 1488 cl->buffer = PSCHED_TICKS2NS(hopt->buffer);
2047 cl->cbuffer = PSCHED_TICKS2NS(hopt->c 1489 cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer);
2048 1490
2049 sch_tree_unlock(sch); 1491 sch_tree_unlock(sch);
2050 qdisc_put(parent_qdisc); <<
2051 1492
2052 if (warn) 1493 if (warn)
2053 NL_SET_ERR_MSG_FMT_MOD(extack !! 1494 pr_warn("HTB: quantum of class %X is %s. Consider r2q change.\n",
2054 "quant !! 1495 cl->common.classid, (warn == -1 ? "small" : "big"));
2055 cl->co <<
2056 1496
2057 qdisc_class_hash_grow(sch, &q->clhash 1497 qdisc_class_hash_grow(sch, &q->clhash);
2058 1498
2059 *arg = (unsigned long)cl; 1499 *arg = (unsigned long)cl;
2060 return 0; 1500 return 0;
2061 1501
2062 err_kill_estimator: <<
2063 gen_kill_estimator(&cl->rate_est); <<
2064 err_block_put: <<
2065 tcf_block_put(cl->block); <<
2066 kfree(cl); <<
2067 failure: 1502 failure:
2068 return err; 1503 return err;
2069 } 1504 }
2070 1505
2071 static struct tcf_block *htb_tcf_block(struct 1506 static struct tcf_block *htb_tcf_block(struct Qdisc *sch, unsigned long arg,
2072 struct 1507 struct netlink_ext_ack *extack)
2073 { 1508 {
2074 struct htb_sched *q = qdisc_priv(sch) 1509 struct htb_sched *q = qdisc_priv(sch);
2075 struct htb_class *cl = (struct htb_cl 1510 struct htb_class *cl = (struct htb_class *)arg;
2076 1511
2077 return cl ? cl->block : q->block; 1512 return cl ? cl->block : q->block;
2078 } 1513 }
2079 1514
2080 static unsigned long htb_bind_filter(struct Q 1515 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
2081 u32 clas 1516 u32 classid)
2082 { 1517 {
2083 struct htb_class *cl = htb_find(class 1518 struct htb_class *cl = htb_find(classid, sch);
2084 1519
2085 /*if (cl && !cl->level) return 0; 1520 /*if (cl && !cl->level) return 0;
2086 * The line above used to be there to 1521 * The line above used to be there to prevent attaching filters to
2087 * leaves. But at least tc_index filt 1522 * leaves. But at least tc_index filter uses this just to get class
2088 * for other reasons so that we have 1523 * for other reasons so that we have to allow for it.
2089 * ---- 1524 * ----
2090 * 19.6.2002 As Werner explained it i 1525 * 19.6.2002 As Werner explained it is ok - bind filter is just
2091 * another way to "lock" the class - 1526 * another way to "lock" the class - unlike "get" this lock can
2092 * be broken by class during destroy 1527 * be broken by class during destroy IIUC.
2093 */ 1528 */
2094 if (cl) 1529 if (cl)
2095 qdisc_class_get(&cl->common); !! 1530 cl->filter_cnt++;
2096 return (unsigned long)cl; 1531 return (unsigned long)cl;
2097 } 1532 }
2098 1533
2099 static void htb_unbind_filter(struct Qdisc *s 1534 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
2100 { 1535 {
2101 struct htb_class *cl = (struct htb_cl 1536 struct htb_class *cl = (struct htb_class *)arg;
2102 1537
2103 qdisc_class_put(&cl->common); !! 1538 if (cl)
>> 1539 cl->filter_cnt--;
2104 } 1540 }
2105 1541
2106 static void htb_walk(struct Qdisc *sch, struc 1542 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
2107 { 1543 {
2108 struct htb_sched *q = qdisc_priv(sch) 1544 struct htb_sched *q = qdisc_priv(sch);
2109 struct htb_class *cl; 1545 struct htb_class *cl;
2110 unsigned int i; 1546 unsigned int i;
2111 1547
2112 if (arg->stop) 1548 if (arg->stop)
2113 return; 1549 return;
2114 1550
2115 for (i = 0; i < q->clhash.hashsize; i 1551 for (i = 0; i < q->clhash.hashsize; i++) {
2116 hlist_for_each_entry(cl, &q-> 1552 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
2117 if (!tc_qdisc_stats_d !! 1553 if (arg->count < arg->skip) {
>> 1554 arg->count++;
>> 1555 continue;
>> 1556 }
>> 1557 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
>> 1558 arg->stop = 1;
2118 return; 1559 return;
>> 1560 }
>> 1561 arg->count++;
2119 } 1562 }
2120 } 1563 }
2121 } 1564 }
2122 1565
2123 static const struct Qdisc_class_ops htb_class 1566 static const struct Qdisc_class_ops htb_class_ops = {
2124 .select_queue = htb_select_qu <<
2125 .graft = htb_graft, 1567 .graft = htb_graft,
2126 .leaf = htb_leaf, 1568 .leaf = htb_leaf,
2127 .qlen_notify = htb_qlen_noti 1569 .qlen_notify = htb_qlen_notify,
2128 .find = htb_search, 1570 .find = htb_search,
2129 .change = htb_change_cl 1571 .change = htb_change_class,
2130 .delete = htb_delete, 1572 .delete = htb_delete,
2131 .walk = htb_walk, 1573 .walk = htb_walk,
2132 .tcf_block = htb_tcf_block 1574 .tcf_block = htb_tcf_block,
2133 .bind_tcf = htb_bind_filt 1575 .bind_tcf = htb_bind_filter,
2134 .unbind_tcf = htb_unbind_fi 1576 .unbind_tcf = htb_unbind_filter,
2135 .dump = htb_dump_clas 1577 .dump = htb_dump_class,
2136 .dump_stats = htb_dump_clas 1578 .dump_stats = htb_dump_class_stats,
2137 }; 1579 };
2138 1580
2139 static struct Qdisc_ops htb_qdisc_ops __read_ 1581 static struct Qdisc_ops htb_qdisc_ops __read_mostly = {
2140 .cl_ops = &htb_class_op 1582 .cl_ops = &htb_class_ops,
2141 .id = "htb", 1583 .id = "htb",
2142 .priv_size = sizeof(struct 1584 .priv_size = sizeof(struct htb_sched),
2143 .enqueue = htb_enqueue, 1585 .enqueue = htb_enqueue,
2144 .dequeue = htb_dequeue, 1586 .dequeue = htb_dequeue,
2145 .peek = qdisc_peek_de 1587 .peek = qdisc_peek_dequeued,
2146 .init = htb_init, 1588 .init = htb_init,
2147 .attach = htb_attach, <<
2148 .reset = htb_reset, 1589 .reset = htb_reset,
2149 .destroy = htb_destroy, 1590 .destroy = htb_destroy,
2150 .dump = htb_dump, 1591 .dump = htb_dump,
2151 .owner = THIS_MODULE, 1592 .owner = THIS_MODULE,
2152 }; 1593 };
2153 MODULE_ALIAS_NET_SCH("htb"); <<
2154 1594
2155 static int __init htb_module_init(void) 1595 static int __init htb_module_init(void)
2156 { 1596 {
2157 return register_qdisc(&htb_qdisc_ops) 1597 return register_qdisc(&htb_qdisc_ops);
2158 } 1598 }
2159 static void __exit htb_module_exit(void) 1599 static void __exit htb_module_exit(void)
2160 { 1600 {
2161 unregister_qdisc(&htb_qdisc_ops); 1601 unregister_qdisc(&htb_qdisc_ops);
2162 } 1602 }
2163 1603
2164 module_init(htb_module_init) 1604 module_init(htb_module_init)
2165 module_exit(htb_module_exit) 1605 module_exit(htb_module_exit)
2166 MODULE_LICENSE("GPL"); 1606 MODULE_LICENSE("GPL");
2167 MODULE_DESCRIPTION("Hierarchical Token Bucket <<
2168 1607
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.