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 {
>> 129 struct list_head drop_list;
125 int defici 130 int deficit[TC_HTB_MAXDEPTH];
126 struct Qdisc *q; 131 struct Qdisc *q;
127 struct netdev_queue *o <<
128 } leaf; 132 } leaf;
129 struct htb_class_inner { 133 struct htb_class_inner {
130 struct htb_prio clprio 134 struct htb_prio clprio[TC_HTB_NUMPRIO];
131 } inner; 135 } inner;
132 }; !! 136 } un;
133 s64 pq_key; 137 s64 pq_key;
134 138
135 int prio_activity; 139 int prio_activity; /* for which prios are we active */
136 enum htb_cmode cmode; 140 enum htb_cmode cmode; /* current mode of the class */
137 struct rb_node pq_node; 141 struct rb_node pq_node; /* node for event queue */
138 struct rb_node node[TC_HTB_NU 142 struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */
139 143
140 unsigned int drops ____cacheline_align 144 unsigned int drops ____cacheline_aligned_in_smp;
141 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 */
>> 173 struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */
171 174
172 /* time of nearest event per level (ro 175 /* time of nearest event per level (row) */
173 s64 near_ev_cache[ 176 s64 near_ev_cache[TC_HTB_MAXDEPTH];
174 177
175 int row_mask[TC_HT 178 int row_mask[TC_HTB_MAXDEPTH];
176 179
177 struct htb_level hlevel[TC_HTB_ 180 struct htb_level hlevel[TC_HTB_MAXDEPTH];
178 <<
179 struct Qdisc **direct_qdisc <<
180 unsigned int num_direct_qdi <<
181 <<
182 bool offload; <<
183 }; 181 };
184 182
185 /* find class in global hash table using given 183 /* find class in global hash table using given handle */
186 static inline struct htb_class *htb_find(u32 h 184 static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch)
187 { 185 {
188 struct htb_sched *q = qdisc_priv(sch); 186 struct htb_sched *q = qdisc_priv(sch);
189 struct Qdisc_class_common *clc; 187 struct Qdisc_class_common *clc;
190 188
191 clc = qdisc_class_find(&q->clhash, han 189 clc = qdisc_class_find(&q->clhash, handle);
192 if (clc == NULL) 190 if (clc == NULL)
193 return NULL; 191 return NULL;
194 return container_of(clc, struct htb_cl 192 return container_of(clc, struct htb_class, common);
195 } 193 }
196 194
197 static unsigned long htb_search(struct Qdisc * 195 static unsigned long htb_search(struct Qdisc *sch, u32 handle)
198 { 196 {
199 return (unsigned long)htb_find(handle, 197 return (unsigned long)htb_find(handle, sch);
200 } 198 }
201 <<
202 #define HTB_DIRECT ((struct htb_class *)-1L) <<
203 <<
204 /** 199 /**
205 * htb_classify - classify a packet into class 200 * 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 * 201 *
210 * It returns NULL if the packet should be dro 202 * It returns NULL if the packet should be dropped or -1 if the packet
211 * should be passed directly thru. In all othe 203 * should be passed directly thru. In all other cases leaf class is returned.
212 * We allow direct class selection by classid 204 * We allow direct class selection by classid in priority. The we examine
213 * filters in qdisc and in inner nodes (if hig 205 * 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 206 * node). If we end up with classid MAJOR:0 we enqueue the skb into special
215 * internal fifo (direct). These packets then 207 * internal fifo (direct). These packets then go directly thru. If we still
216 * have no valid leaf we try to use MAJOR:defa 208 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful
217 * then finish and return direct queue. 209 * then finish and return direct queue.
218 */ 210 */
>> 211 #define HTB_DIRECT ((struct htb_class *)-1L)
>> 212
219 static struct htb_class *htb_classify(struct s 213 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch,
220 int *qer 214 int *qerr)
221 { 215 {
222 struct htb_sched *q = qdisc_priv(sch); 216 struct htb_sched *q = qdisc_priv(sch);
223 struct htb_class *cl; 217 struct htb_class *cl;
224 struct tcf_result res; 218 struct tcf_result res;
225 struct tcf_proto *tcf; 219 struct tcf_proto *tcf;
226 int result; 220 int result;
227 221
228 /* allow to select class by setting sk 222 /* allow to select class by setting skb->priority to valid classid;
229 * note that nfmark can be used too by 223 * note that nfmark can be used too by attaching filter fw with no
230 * rules in it 224 * rules in it
231 */ 225 */
232 if (skb->priority == sch->handle) 226 if (skb->priority == sch->handle)
233 return HTB_DIRECT; /* X:0 227 return HTB_DIRECT; /* X:0 (direct flow) selected */
234 cl = htb_find(skb->priority, sch); 228 cl = htb_find(skb->priority, sch);
235 if (cl) { 229 if (cl) {
236 if (cl->level == 0) 230 if (cl->level == 0)
237 return cl; 231 return cl;
238 /* Start with inner filter cha 232 /* Start with inner filter chain if a non-leaf class is selected */
239 tcf = rcu_dereference_bh(cl->f 233 tcf = rcu_dereference_bh(cl->filter_list);
240 } else { 234 } else {
241 tcf = rcu_dereference_bh(q->fi 235 tcf = rcu_dereference_bh(q->filter_list);
242 } 236 }
243 237
244 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_ 238 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
245 while (tcf && (result = tcf_classify(s !! 239 while (tcf && (result = tcf_classify(skb, tcf, &res, false)) >= 0) {
246 #ifdef CONFIG_NET_CLS_ACT 240 #ifdef CONFIG_NET_CLS_ACT
247 switch (result) { 241 switch (result) {
248 case TC_ACT_QUEUED: 242 case TC_ACT_QUEUED:
249 case TC_ACT_STOLEN: 243 case TC_ACT_STOLEN:
250 case TC_ACT_TRAP: 244 case TC_ACT_TRAP:
251 *qerr = NET_XMIT_SUCCE 245 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
252 fallthrough; <<
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->un.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->un.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->un.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->un.inner.clprio[prio].last_ptr_id = cl->common.classid;
483 p->inner.clpri !! 456 p->un.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->un.inner.clprio[prio].feed);
488 461
489 if (!p->inner.clprio[p !! 462 if (!p->un.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) { <<
569 cl->overlimits++; <<
570 q->overlimits++; <<
571 } <<
572 <<
573 if (cl->prio_activity) { /* not 536 if (cl->prio_activity) { /* not necessary: speed optimization */
574 if (cl->cmode != HTB_CANT_SEND 537 if (cl->cmode != HTB_CANT_SEND)
575 htb_deactivate_prios(q 538 htb_deactivate_prios(q, cl);
576 cl->cmode = new_mode; 539 cl->cmode = new_mode;
577 if (new_mode != HTB_CANT_SEND) 540 if (new_mode != HTB_CANT_SEND)
578 htb_activate_prios(q, 541 htb_activate_prios(q, cl);
579 } else 542 } else
580 cl->cmode = new_mode; 543 cl->cmode = new_mode;
581 } 544 }
582 545
583 /** 546 /**
584 * htb_activate - inserts leaf cl into appropr 547 * htb_activate - inserts leaf cl into appropriate active feeds
585 * @q: the priority event queue <<
586 * @cl: the target class <<
587 * 548 *
588 * Routine learns (new) priority of leaf and a 549 * Routine learns (new) priority of leaf and activates feed chain
589 * for the prio. It can be called on already a 550 * for the prio. It can be called on already active leaf safely.
590 * It also adds leaf into droplist. 551 * It also adds leaf into droplist.
591 */ 552 */
592 static inline void htb_activate(struct htb_sch 553 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl)
593 { 554 {
594 WARN_ON(cl->level || !cl->leaf.q || !c !! 555 WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen);
595 556
596 if (!cl->prio_activity) { 557 if (!cl->prio_activity) {
597 cl->prio_activity = 1 << cl->p 558 cl->prio_activity = 1 << cl->prio;
598 htb_activate_prios(q, cl); 559 htb_activate_prios(q, cl);
>> 560 list_add_tail(&cl->un.leaf.drop_list,
>> 561 q->drops + cl->prio);
599 } 562 }
600 } 563 }
601 564
602 /** 565 /**
603 * htb_deactivate - remove leaf cl from active 566 * htb_deactivate - remove leaf cl from active feeds
604 * @q: the priority event queue <<
605 * @cl: the target class <<
606 * 567 *
607 * Make sure that leaf is active. In the other 568 * Make sure that leaf is active. In the other words it can't be called
608 * with non-active leaf. It also removes class 569 * with non-active leaf. It also removes class from the drop list.
609 */ 570 */
610 static inline void htb_deactivate(struct htb_s 571 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl)
611 { 572 {
612 WARN_ON(!cl->prio_activity); 573 WARN_ON(!cl->prio_activity);
613 574
614 htb_deactivate_prios(q, cl); 575 htb_deactivate_prios(q, cl);
615 cl->prio_activity = 0; 576 cl->prio_activity = 0;
>> 577 list_del_init(&cl->un.leaf.drop_list);
>> 578 }
>> 579
>> 580 static void htb_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
>> 581 struct qdisc_skb_head *qh)
>> 582 {
>> 583 struct sk_buff *last = qh->tail;
>> 584
>> 585 if (last) {
>> 586 skb->next = NULL;
>> 587 last->next = skb;
>> 588 qh->tail = skb;
>> 589 } else {
>> 590 qh->tail = skb;
>> 591 qh->head = skb;
>> 592 }
>> 593 qh->qlen++;
616 } 594 }
617 595
618 static int htb_enqueue(struct sk_buff *skb, st 596 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch,
619 struct sk_buff **to_fre 597 struct sk_buff **to_free)
620 { 598 {
621 int ret; !! 599 int uninitialized_var(ret);
622 unsigned int len = qdisc_pkt_len(skb); <<
623 struct htb_sched *q = qdisc_priv(sch); 600 struct htb_sched *q = qdisc_priv(sch);
624 struct htb_class *cl = htb_classify(sk 601 struct htb_class *cl = htb_classify(skb, sch, &ret);
625 602
626 if (cl == HTB_DIRECT) { 603 if (cl == HTB_DIRECT) {
627 /* enqueue to helper queue */ 604 /* enqueue to helper queue */
628 if (q->direct_queue.qlen < q-> 605 if (q->direct_queue.qlen < q->direct_qlen) {
629 __qdisc_enqueue_tail(s !! 606 htb_enqueue_tail(skb, sch, &q->direct_queue);
630 q->direct_pkts++; 607 q->direct_pkts++;
631 } else { 608 } else {
632 return qdisc_drop(skb, 609 return qdisc_drop(skb, sch, to_free);
633 } 610 }
634 #ifdef CONFIG_NET_CLS_ACT 611 #ifdef CONFIG_NET_CLS_ACT
635 } else if (!cl) { 612 } else if (!cl) {
636 if (ret & __NET_XMIT_BYPASS) 613 if (ret & __NET_XMIT_BYPASS)
637 qdisc_qstats_drop(sch) 614 qdisc_qstats_drop(sch);
638 __qdisc_drop(skb, to_free); 615 __qdisc_drop(skb, to_free);
639 return ret; 616 return ret;
640 #endif 617 #endif
641 } else if ((ret = qdisc_enqueue(skb, c !! 618 } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q,
642 to_fre 619 to_free)) != NET_XMIT_SUCCESS) {
643 if (net_xmit_drop_count(ret)) 620 if (net_xmit_drop_count(ret)) {
644 qdisc_qstats_drop(sch) 621 qdisc_qstats_drop(sch);
645 cl->drops++; 622 cl->drops++;
646 } 623 }
647 return ret; 624 return ret;
648 } else { 625 } else {
649 htb_activate(q, cl); 626 htb_activate(q, cl);
650 } 627 }
651 628
652 sch->qstats.backlog += len; !! 629 qdisc_qstats_backlog_inc(sch, skb);
653 sch->q.qlen++; 630 sch->q.qlen++;
654 return NET_XMIT_SUCCESS; 631 return NET_XMIT_SUCCESS;
655 } 632 }
656 633
657 static inline void htb_accnt_tokens(struct htb 634 static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, s64 diff)
658 { 635 {
659 s64 toks = diff + cl->tokens; 636 s64 toks = diff + cl->tokens;
660 637
661 if (toks > cl->buffer) 638 if (toks > cl->buffer)
662 toks = cl->buffer; 639 toks = cl->buffer;
663 toks -= (s64) psched_l2t_ns(&cl->rate, 640 toks -= (s64) psched_l2t_ns(&cl->rate, bytes);
664 if (toks <= -cl->mbuffer) 641 if (toks <= -cl->mbuffer)
665 toks = 1 - cl->mbuffer; 642 toks = 1 - cl->mbuffer;
666 643
667 cl->tokens = toks; 644 cl->tokens = toks;
668 } 645 }
669 646
670 static inline void htb_accnt_ctokens(struct ht 647 static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, s64 diff)
671 { 648 {
672 s64 toks = diff + cl->ctokens; 649 s64 toks = diff + cl->ctokens;
673 650
674 if (toks > cl->cbuffer) 651 if (toks > cl->cbuffer)
675 toks = cl->cbuffer; 652 toks = cl->cbuffer;
676 toks -= (s64) psched_l2t_ns(&cl->ceil, 653 toks -= (s64) psched_l2t_ns(&cl->ceil, bytes);
677 if (toks <= -cl->mbuffer) 654 if (toks <= -cl->mbuffer)
678 toks = 1 - cl->mbuffer; 655 toks = 1 - cl->mbuffer;
679 656
680 cl->ctokens = toks; 657 cl->ctokens = toks;
681 } 658 }
682 659
683 /** 660 /**
684 * htb_charge_class - charges amount "bytes" t 661 * 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 * 662 *
690 * Routine assumes that packet "bytes" long wa 663 * Routine assumes that packet "bytes" long was dequeued from leaf cl
691 * borrowing from "level". It accounts bytes t 664 * borrowing from "level". It accounts bytes to ceil leaky bucket for
692 * leaf and all ancestors and to rate bucket f 665 * leaf and all ancestors and to rate bucket for ancestors at levels
693 * "level" and higher. It also handles possibl 666 * "level" and higher. It also handles possible change of mode resulting
694 * from the update. Note that mode can also in 667 * from the update. Note that mode can also increase here (MAY_BORROW to
695 * CAN_SEND) because we can use more precise c 668 * CAN_SEND) because we can use more precise clock that event queue here.
696 * In such case we remove class from event que 669 * In such case we remove class from event queue first.
697 */ 670 */
698 static void htb_charge_class(struct htb_sched 671 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl,
699 int level, struct 672 int level, struct sk_buff *skb)
700 { 673 {
701 int bytes = qdisc_pkt_len(skb); 674 int bytes = qdisc_pkt_len(skb);
702 enum htb_cmode old_mode; 675 enum htb_cmode old_mode;
703 s64 diff; 676 s64 diff;
704 677
705 while (cl) { 678 while (cl) {
706 diff = min_t(s64, q->now - cl- 679 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer);
707 if (cl->level >= level) { 680 if (cl->level >= level) {
708 if (cl->level == level 681 if (cl->level == level)
709 cl->xstats.len 682 cl->xstats.lends++;
710 htb_accnt_tokens(cl, b 683 htb_accnt_tokens(cl, bytes, diff);
711 } else { 684 } else {
712 cl->xstats.borrows++; 685 cl->xstats.borrows++;
713 cl->tokens += diff; 686 cl->tokens += diff; /* we moved t_c; update tokens */
714 } 687 }
715 htb_accnt_ctokens(cl, bytes, d 688 htb_accnt_ctokens(cl, bytes, diff);
716 cl->t_c = q->now; 689 cl->t_c = q->now;
717 690
718 old_mode = cl->cmode; 691 old_mode = cl->cmode;
719 diff = 0; 692 diff = 0;
720 htb_change_class_mode(q, cl, & 693 htb_change_class_mode(q, cl, &diff);
721 if (old_mode != cl->cmode) { 694 if (old_mode != cl->cmode) {
722 if (old_mode != HTB_CA 695 if (old_mode != HTB_CAN_SEND)
723 htb_safe_rb_er 696 htb_safe_rb_erase(&cl->pq_node, &q->hlevel[cl->level].wait_pq);
724 if (cl->cmode != HTB_C 697 if (cl->cmode != HTB_CAN_SEND)
725 htb_add_to_wai 698 htb_add_to_wait_tree(q, cl, diff);
726 } 699 }
727 700
728 /* update basic stats except f 701 /* update basic stats except for leaves which are already updated */
729 if (cl->level) 702 if (cl->level)
730 bstats_update(&cl->bst 703 bstats_update(&cl->bstats, skb);
731 704
732 cl = cl->parent; 705 cl = cl->parent;
733 } 706 }
734 } 707 }
735 708
736 /** 709 /**
737 * htb_do_events - make mode changes to classe 710 * 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 * 711 *
742 * Scans event queue for pending events and ap 712 * Scans event queue for pending events and applies them. Returns time of
743 * next pending event (0 for no event in pq, q 713 * next pending event (0 for no event in pq, q->now for too many events).
744 * Note: Applied are events whose have cl->pq_ 714 * Note: Applied are events whose have cl->pq_key <= q->now.
745 */ 715 */
746 static s64 htb_do_events(struct htb_sched *q, 716 static s64 htb_do_events(struct htb_sched *q, const int level,
747 unsigned long start) 717 unsigned long start)
748 { 718 {
749 /* don't run for longer than 2 jiffies 719 /* don't run for longer than 2 jiffies; 2 is used instead of
750 * 1 to simplify things when jiffy is 720 * 1 to simplify things when jiffy is going to be incremented
751 * too soon 721 * too soon
752 */ 722 */
753 unsigned long stop_at = start + 2; 723 unsigned long stop_at = start + 2;
754 struct rb_root *wait_pq = &q->hlevel[l 724 struct rb_root *wait_pq = &q->hlevel[level].wait_pq;
755 725
756 while (time_before(jiffies, stop_at)) 726 while (time_before(jiffies, stop_at)) {
757 struct htb_class *cl; 727 struct htb_class *cl;
758 s64 diff; 728 s64 diff;
759 struct rb_node *p = rb_first(w 729 struct rb_node *p = rb_first(wait_pq);
760 730
761 if (!p) 731 if (!p)
762 return 0; 732 return 0;
763 733
764 cl = rb_entry(p, struct htb_cl 734 cl = rb_entry(p, struct htb_class, pq_node);
765 if (cl->pq_key > q->now) 735 if (cl->pq_key > q->now)
766 return cl->pq_key; 736 return cl->pq_key;
767 737
768 htb_safe_rb_erase(p, wait_pq); 738 htb_safe_rb_erase(p, wait_pq);
769 diff = min_t(s64, q->now - cl- 739 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer);
770 htb_change_class_mode(q, cl, & 740 htb_change_class_mode(q, cl, &diff);
771 if (cl->cmode != HTB_CAN_SEND) 741 if (cl->cmode != HTB_CAN_SEND)
772 htb_add_to_wait_tree(q 742 htb_add_to_wait_tree(q, cl, diff);
773 } 743 }
774 744
775 /* too much load - let's continue afte 745 /* too much load - let's continue after a break for scheduling */
776 if (!(q->warned & HTB_WARN_TOOMANYEVEN 746 if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) {
777 pr_warn("htb: too many events! 747 pr_warn("htb: too many events!\n");
778 q->warned |= HTB_WARN_TOOMANYE 748 q->warned |= HTB_WARN_TOOMANYEVENTS;
779 } 749 }
780 750
781 return q->now; 751 return q->now;
782 } 752 }
783 753
784 /* Returns class->node+prio from id-tree where 754 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
785 * is no such one exists. 755 * is no such one exists.
786 */ 756 */
787 static struct rb_node *htb_id_find_next_upper( 757 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n,
788 758 u32 id)
789 { 759 {
790 struct rb_node *r = NULL; 760 struct rb_node *r = NULL;
791 while (n) { 761 while (n) {
792 struct htb_class *cl = 762 struct htb_class *cl =
793 rb_entry(n, struct htb_cla 763 rb_entry(n, struct htb_class, node[prio]);
794 764
795 if (id > cl->common.classid) { 765 if (id > cl->common.classid) {
796 n = n->rb_right; 766 n = n->rb_right;
797 } else if (id < cl->common.cla 767 } else if (id < cl->common.classid) {
798 r = n; 768 r = n;
799 n = n->rb_left; 769 n = n->rb_left;
800 } else { 770 } else {
801 return n; 771 return n;
802 } 772 }
803 } 773 }
804 return r; 774 return r;
805 } 775 }
806 776
807 /** 777 /**
808 * htb_lookup_leaf - returns next leaf class i 778 * htb_lookup_leaf - returns next leaf class in DRR order
809 * @hprio: the current one <<
810 * @prio: which prio in class <<
811 * 779 *
812 * Find leaf where current feed pointers point 780 * Find leaf where current feed pointers points to.
813 */ 781 */
814 static struct htb_class *htb_lookup_leaf(struc 782 static struct htb_class *htb_lookup_leaf(struct htb_prio *hprio, const int prio)
815 { 783 {
816 int i; 784 int i;
817 struct { 785 struct {
818 struct rb_node *root; 786 struct rb_node *root;
819 struct rb_node **pptr; 787 struct rb_node **pptr;
820 u32 *pid; 788 u32 *pid;
821 } stk[TC_HTB_MAXDEPTH], *sp = stk; 789 } stk[TC_HTB_MAXDEPTH], *sp = stk;
822 790
823 BUG_ON(!hprio->row.rb_node); 791 BUG_ON(!hprio->row.rb_node);
824 sp->root = hprio->row.rb_node; 792 sp->root = hprio->row.rb_node;
825 sp->pptr = &hprio->ptr; 793 sp->pptr = &hprio->ptr;
826 sp->pid = &hprio->last_ptr_id; 794 sp->pid = &hprio->last_ptr_id;
827 795
828 for (i = 0; i < 65535; i++) { 796 for (i = 0; i < 65535; i++) {
829 if (!*sp->pptr && *sp->pid) { 797 if (!*sp->pptr && *sp->pid) {
830 /* ptr was invalidated 798 /* ptr was invalidated but id is valid - try to recover
831 * the original or nex 799 * the original or next ptr
832 */ 800 */
833 *sp->pptr = 801 *sp->pptr =
834 htb_id_find_next_u 802 htb_id_find_next_upper(prio, sp->root, *sp->pid);
835 } 803 }
836 *sp->pid = 0; /* ptr is vali 804 *sp->pid = 0; /* ptr is valid now so that remove this hint as it
837 * can become 805 * can become out of date quickly
838 */ 806 */
839 if (!*sp->pptr) { /* we 807 if (!*sp->pptr) { /* we are at right end; rewind & go up */
840 *sp->pptr = sp->root; 808 *sp->pptr = sp->root;
841 while ((*sp->pptr)->rb 809 while ((*sp->pptr)->rb_left)
842 *sp->pptr = (* 810 *sp->pptr = (*sp->pptr)->rb_left;
843 if (sp > stk) { 811 if (sp > stk) {
844 sp--; 812 sp--;
845 if (!*sp->pptr 813 if (!*sp->pptr) {
846 WARN_O 814 WARN_ON(1);
847 return 815 return NULL;
848 } 816 }
849 htb_next_rb_no 817 htb_next_rb_node(sp->pptr);
850 } 818 }
851 } else { 819 } else {
852 struct htb_class *cl; 820 struct htb_class *cl;
853 struct htb_prio *clp; 821 struct htb_prio *clp;
854 822
855 cl = rb_entry(*sp->ppt 823 cl = rb_entry(*sp->pptr, struct htb_class, node[prio]);
856 if (!cl->level) 824 if (!cl->level)
857 return cl; 825 return cl;
858 clp = &cl->inner.clpri !! 826 clp = &cl->un.inner.clprio[prio];
859 (++sp)->root = clp->fe 827 (++sp)->root = clp->feed.rb_node;
860 sp->pptr = &clp->ptr; 828 sp->pptr = &clp->ptr;
861 sp->pid = &clp->last_p 829 sp->pid = &clp->last_ptr_id;
862 } 830 }
863 } 831 }
864 WARN_ON(1); 832 WARN_ON(1);
865 return NULL; 833 return NULL;
866 } 834 }
867 835
868 /* dequeues packet at given priority and level 836 /* dequeues packet at given priority and level; call only if
869 * you are sure that there is active class at 837 * you are sure that there is active class at prio/level
870 */ 838 */
871 static struct sk_buff *htb_dequeue_tree(struct 839 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, const int prio,
872 const 840 const int level)
873 { 841 {
874 struct sk_buff *skb = NULL; 842 struct sk_buff *skb = NULL;
875 struct htb_class *cl, *start; 843 struct htb_class *cl, *start;
876 struct htb_level *hlevel = &q->hlevel[ 844 struct htb_level *hlevel = &q->hlevel[level];
877 struct htb_prio *hprio = &hlevel->hpri 845 struct htb_prio *hprio = &hlevel->hprio[prio];
878 846
879 /* look initial class up in the row */ 847 /* look initial class up in the row */
880 start = cl = htb_lookup_leaf(hprio, pr 848 start = cl = htb_lookup_leaf(hprio, prio);
881 849
882 do { 850 do {
883 next: 851 next:
884 if (unlikely(!cl)) 852 if (unlikely(!cl))
885 return NULL; 853 return NULL;
886 854
887 /* class can be empty - it is 855 /* class can be empty - it is unlikely but can be true if leaf
888 * qdisc drops packets in enqu 856 * qdisc drops packets in enqueue routine or if someone used
889 * graft operation on the leaf 857 * graft operation on the leaf since last dequeue;
890 * simply deactivate and skip 858 * simply deactivate and skip such class
891 */ 859 */
892 if (unlikely(cl->leaf.q->q.qle !! 860 if (unlikely(cl->un.leaf.q->q.qlen == 0)) {
893 struct htb_class *next 861 struct htb_class *next;
894 htb_deactivate(q, cl); 862 htb_deactivate(q, cl);
895 863
896 /* row/level might bec 864 /* row/level might become empty */
897 if ((q->row_mask[level 865 if ((q->row_mask[level] & (1 << prio)) == 0)
898 return NULL; 866 return NULL;
899 867
900 next = htb_lookup_leaf 868 next = htb_lookup_leaf(hprio, prio);
901 869
902 if (cl == start) 870 if (cl == start) /* fix start if we just deleted it */
903 start = next; 871 start = next;
904 cl = next; 872 cl = next;
905 goto next; 873 goto next;
906 } 874 }
907 875
908 skb = cl->leaf.q->dequeue(cl-> !! 876 skb = cl->un.leaf.q->dequeue(cl->un.leaf.q);
909 if (likely(skb != NULL)) 877 if (likely(skb != NULL))
910 break; 878 break;
911 879
912 qdisc_warn_nonwc("htb", cl->le !! 880 qdisc_warn_nonwc("htb", cl->un.leaf.q);
913 htb_next_rb_node(level ? &cl-> !! 881 htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr:
914 &q->h 882 &q->hlevel[0].hprio[prio].ptr);
915 cl = htb_lookup_leaf(hprio, pr 883 cl = htb_lookup_leaf(hprio, prio);
916 884
917 } while (cl != start); 885 } while (cl != start);
918 886
919 if (likely(skb != NULL)) { 887 if (likely(skb != NULL)) {
920 bstats_update(&cl->bstats, skb 888 bstats_update(&cl->bstats, skb);
921 cl->leaf.deficit[level] -= qdi !! 889 cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb);
922 if (cl->leaf.deficit[level] < !! 890 if (cl->un.leaf.deficit[level] < 0) {
923 cl->leaf.deficit[level !! 891 cl->un.leaf.deficit[level] += cl->quantum;
924 htb_next_rb_node(level !! 892 htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr :
925 893 &q->hlevel[0].hprio[prio].ptr);
926 } 894 }
927 /* this used to be after charg 895 /* this used to be after charge_class but this constelation
928 * gives us slightly better pe 896 * gives us slightly better performance
929 */ 897 */
930 if (!cl->leaf.q->q.qlen) !! 898 if (!cl->un.leaf.q->q.qlen)
931 htb_deactivate(q, cl); 899 htb_deactivate(q, cl);
932 htb_charge_class(q, cl, level, 900 htb_charge_class(q, cl, level, skb);
933 } 901 }
934 return skb; 902 return skb;
935 } 903 }
936 904
937 static struct sk_buff *htb_dequeue(struct Qdis 905 static struct sk_buff *htb_dequeue(struct Qdisc *sch)
938 { 906 {
939 struct sk_buff *skb; 907 struct sk_buff *skb;
940 struct htb_sched *q = qdisc_priv(sch); 908 struct htb_sched *q = qdisc_priv(sch);
941 int level; 909 int level;
942 s64 next_event; 910 s64 next_event;
943 unsigned long start_at; 911 unsigned long start_at;
944 912
945 /* try to dequeue direct packets as hi 913 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
946 skb = __qdisc_dequeue_head(&q->direct_ 914 skb = __qdisc_dequeue_head(&q->direct_queue);
947 if (skb != NULL) { 915 if (skb != NULL) {
948 ok: 916 ok:
949 qdisc_bstats_update(sch, skb); 917 qdisc_bstats_update(sch, skb);
950 qdisc_qstats_backlog_dec(sch, 918 qdisc_qstats_backlog_dec(sch, skb);
951 sch->q.qlen--; 919 sch->q.qlen--;
952 return skb; 920 return skb;
953 } 921 }
954 922
955 if (!sch->q.qlen) 923 if (!sch->q.qlen)
956 goto fin; 924 goto fin;
957 q->now = ktime_get_ns(); 925 q->now = ktime_get_ns();
958 start_at = jiffies; 926 start_at = jiffies;
959 927
960 next_event = q->now + 5LLU * NSEC_PER_ 928 next_event = q->now + 5LLU * NSEC_PER_SEC;
961 929
962 for (level = 0; level < TC_HTB_MAXDEPT 930 for (level = 0; level < TC_HTB_MAXDEPTH; level++) {
963 /* common case optimization - 931 /* common case optimization - skip event handler quickly */
964 int m; 932 int m;
965 s64 event = q->near_ev_cache[l 933 s64 event = q->near_ev_cache[level];
966 934
967 if (q->now >= event) { 935 if (q->now >= event) {
968 event = htb_do_events( 936 event = htb_do_events(q, level, start_at);
969 if (!event) 937 if (!event)
970 event = q->now 938 event = q->now + NSEC_PER_SEC;
971 q->near_ev_cache[level 939 q->near_ev_cache[level] = event;
972 } 940 }
973 941
974 if (next_event > event) 942 if (next_event > event)
975 next_event = event; 943 next_event = event;
976 944
977 m = ~q->row_mask[level]; 945 m = ~q->row_mask[level];
978 while (m != (int)(-1)) { 946 while (m != (int)(-1)) {
979 int prio = ffz(m); 947 int prio = ffz(m);
980 948
981 m |= 1 << prio; 949 m |= 1 << prio;
982 skb = htb_dequeue_tree 950 skb = htb_dequeue_tree(q, prio, level);
983 if (likely(skb != NULL 951 if (likely(skb != NULL))
984 goto ok; 952 goto ok;
985 } 953 }
986 } 954 }
>> 955 qdisc_qstats_overlimit(sch);
987 if (likely(next_event > q->now)) 956 if (likely(next_event > q->now))
988 qdisc_watchdog_schedule_ns(&q- 957 qdisc_watchdog_schedule_ns(&q->watchdog, next_event);
989 else 958 else
990 schedule_work(&q->work); 959 schedule_work(&q->work);
991 fin: 960 fin:
992 return skb; 961 return skb;
993 } 962 }
994 963
995 /* reset all classes */ 964 /* reset all classes */
996 /* always caled under BH & queue lock */ 965 /* always caled under BH & queue lock */
997 static void htb_reset(struct Qdisc *sch) 966 static void htb_reset(struct Qdisc *sch)
998 { 967 {
999 struct htb_sched *q = qdisc_priv(sch); 968 struct htb_sched *q = qdisc_priv(sch);
1000 struct htb_class *cl; 969 struct htb_class *cl;
1001 unsigned int i; 970 unsigned int i;
1002 971
1003 for (i = 0; i < q->clhash.hashsize; i 972 for (i = 0; i < q->clhash.hashsize; i++) {
1004 hlist_for_each_entry(cl, &q-> 973 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
1005 if (cl->level) 974 if (cl->level)
1006 memset(&cl->i !! 975 memset(&cl->un.inner, 0, sizeof(cl->un.inner));
1007 else { 976 else {
1008 if (cl->leaf. !! 977 if (cl->un.leaf.q)
1009 qdisc !! 978 qdisc_reset(cl->un.leaf.q);
>> 979 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
1010 } 980 }
1011 cl->prio_activity = 0 981 cl->prio_activity = 0;
1012 cl->cmode = HTB_CAN_S 982 cl->cmode = HTB_CAN_SEND;
1013 } 983 }
1014 } 984 }
1015 qdisc_watchdog_cancel(&q->watchdog); 985 qdisc_watchdog_cancel(&q->watchdog);
1016 __qdisc_reset_queue(&q->direct_queue) 986 __qdisc_reset_queue(&q->direct_queue);
>> 987 sch->q.qlen = 0;
>> 988 sch->qstats.backlog = 0;
1017 memset(q->hlevel, 0, sizeof(q->hlevel 989 memset(q->hlevel, 0, sizeof(q->hlevel));
1018 memset(q->row_mask, 0, sizeof(q->row_ 990 memset(q->row_mask, 0, sizeof(q->row_mask));
>> 991 for (i = 0; i < TC_HTB_NUMPRIO; i++)
>> 992 INIT_LIST_HEAD(q->drops + i);
1019 } 993 }
1020 994
1021 static const struct nla_policy htb_policy[TCA 995 static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = {
1022 [TCA_HTB_PARMS] = { .len = sizeof(str 996 [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) },
1023 [TCA_HTB_INIT] = { .len = sizeof(str 997 [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) },
1024 [TCA_HTB_CTAB] = { .type = NLA_BINAR 998 [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1025 [TCA_HTB_RTAB] = { .type = NLA_BINAR 999 [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1026 [TCA_HTB_DIRECT_QLEN] = { .type = NLA 1000 [TCA_HTB_DIRECT_QLEN] = { .type = NLA_U32 },
1027 [TCA_HTB_RATE64] = { .type = NLA_U64 1001 [TCA_HTB_RATE64] = { .type = NLA_U64 },
1028 [TCA_HTB_CEIL64] = { .type = NLA_U64 1002 [TCA_HTB_CEIL64] = { .type = NLA_U64 },
1029 [TCA_HTB_OFFLOAD] = { .type = NLA_FLA <<
1030 }; 1003 };
1031 1004
1032 static void htb_work_func(struct work_struct 1005 static void htb_work_func(struct work_struct *work)
1033 { 1006 {
1034 struct htb_sched *q = container_of(wo 1007 struct htb_sched *q = container_of(work, struct htb_sched, work);
1035 struct Qdisc *sch = q->watchdog.qdisc 1008 struct Qdisc *sch = q->watchdog.qdisc;
1036 1009
1037 rcu_read_lock(); 1010 rcu_read_lock();
1038 __netif_schedule(qdisc_root(sch)); 1011 __netif_schedule(qdisc_root(sch));
1039 rcu_read_unlock(); 1012 rcu_read_unlock();
1040 } 1013 }
1041 1014
1042 static int htb_offload(struct net_device *dev !! 1015 static int htb_init(struct Qdisc *sch, struct nlattr *opt)
1043 { <<
1044 return dev->netdev_ops->ndo_setup_tc( <<
1045 } <<
1046 <<
1047 static int htb_init(struct Qdisc *sch, struct <<
1048 struct netlink_ext_ack *e <<
1049 { 1016 {
1050 struct net_device *dev = qdisc_dev(sc <<
1051 struct tc_htb_qopt_offload offload_op <<
1052 struct htb_sched *q = qdisc_priv(sch) 1017 struct htb_sched *q = qdisc_priv(sch);
1053 struct nlattr *tb[TCA_HTB_MAX + 1]; 1018 struct nlattr *tb[TCA_HTB_MAX + 1];
1054 struct tc_htb_glob *gopt; 1019 struct tc_htb_glob *gopt;
1055 unsigned int ntx; <<
1056 bool offload; <<
1057 int err; 1020 int err;
>> 1021 int i;
1058 1022
1059 qdisc_watchdog_init(&q->watchdog, sch 1023 qdisc_watchdog_init(&q->watchdog, sch);
1060 INIT_WORK(&q->work, htb_work_func); 1024 INIT_WORK(&q->work, htb_work_func);
1061 1025
1062 if (!opt) 1026 if (!opt)
1063 return -EINVAL; 1027 return -EINVAL;
1064 1028
1065 err = tcf_block_get(&q->block, &q->fi !! 1029 err = tcf_block_get(&q->block, &q->filter_list);
1066 if (err) 1030 if (err)
1067 return err; 1031 return err;
1068 1032
1069 err = nla_parse_nested_deprecated(tb, !! 1033 err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy, NULL);
1070 NUL <<
1071 if (err < 0) 1034 if (err < 0)
1072 return err; 1035 return err;
1073 1036
1074 if (!tb[TCA_HTB_INIT]) 1037 if (!tb[TCA_HTB_INIT])
1075 return -EINVAL; 1038 return -EINVAL;
1076 1039
1077 gopt = nla_data(tb[TCA_HTB_INIT]); 1040 gopt = nla_data(tb[TCA_HTB_INIT]);
1078 if (gopt->version != HTB_VER >> 16) 1041 if (gopt->version != HTB_VER >> 16)
1079 return -EINVAL; 1042 return -EINVAL;
1080 1043
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 1044 err = qdisc_class_hash_init(&q->clhash);
1103 if (err < 0) 1045 if (err < 0)
1104 return err; 1046 return err;
>> 1047 for (i = 0; i < TC_HTB_NUMPRIO; i++)
>> 1048 INIT_LIST_HEAD(q->drops + i);
>> 1049
>> 1050 qdisc_skb_head_init(&q->direct_queue);
1105 1051
1106 if (tb[TCA_HTB_DIRECT_QLEN]) 1052 if (tb[TCA_HTB_DIRECT_QLEN])
1107 q->direct_qlen = nla_get_u32( 1053 q->direct_qlen = nla_get_u32(tb[TCA_HTB_DIRECT_QLEN]);
1108 else 1054 else
1109 q->direct_qlen = qdisc_dev(sc 1055 q->direct_qlen = qdisc_dev(sch)->tx_queue_len;
1110 1056
1111 if ((q->rate2quantum = gopt->rate2qua 1057 if ((q->rate2quantum = gopt->rate2quantum) < 1)
1112 q->rate2quantum = 1; 1058 q->rate2quantum = 1;
1113 q->defcls = gopt->defcls; 1059 q->defcls = gopt->defcls;
1114 1060
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; 1061 return 0;
1150 } 1062 }
1151 1063
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 1064 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1203 { 1065 {
1204 struct htb_sched *q = qdisc_priv(sch) 1066 struct htb_sched *q = qdisc_priv(sch);
1205 struct nlattr *nest; 1067 struct nlattr *nest;
1206 struct tc_htb_glob gopt; 1068 struct tc_htb_glob gopt;
1207 1069
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 1070 /* Its safe to not acquire qdisc lock. As we hold RTNL,
1215 * no change can happen on the qdisc 1071 * no change can happen on the qdisc parameters.
1216 */ 1072 */
1217 1073
1218 gopt.direct_pkts = q->direct_pkts; 1074 gopt.direct_pkts = q->direct_pkts;
1219 gopt.version = HTB_VER; 1075 gopt.version = HTB_VER;
1220 gopt.rate2quantum = q->rate2quantum; 1076 gopt.rate2quantum = q->rate2quantum;
1221 gopt.defcls = q->defcls; 1077 gopt.defcls = q->defcls;
1222 gopt.debug = 0; 1078 gopt.debug = 0;
1223 1079
1224 nest = nla_nest_start_noflag(skb, TCA !! 1080 nest = nla_nest_start(skb, TCA_OPTIONS);
1225 if (nest == NULL) 1081 if (nest == NULL)
1226 goto nla_put_failure; 1082 goto nla_put_failure;
1227 if (nla_put(skb, TCA_HTB_INIT, sizeof 1083 if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt) ||
1228 nla_put_u32(skb, TCA_HTB_DIRECT_Q 1084 nla_put_u32(skb, TCA_HTB_DIRECT_QLEN, q->direct_qlen))
1229 goto nla_put_failure; 1085 goto nla_put_failure;
1230 if (q->offload && nla_put_flag(skb, T <<
1231 goto nla_put_failure; <<
1232 1086
1233 return nla_nest_end(skb, nest); 1087 return nla_nest_end(skb, nest);
1234 1088
1235 nla_put_failure: 1089 nla_put_failure:
1236 nla_nest_cancel(skb, nest); 1090 nla_nest_cancel(skb, nest);
1237 return -1; 1091 return -1;
1238 } 1092 }
1239 1093
1240 static int htb_dump_class(struct Qdisc *sch, 1094 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1241 struct sk_buff *skb 1095 struct sk_buff *skb, struct tcmsg *tcm)
1242 { 1096 {
1243 struct htb_class *cl = (struct htb_cl 1097 struct htb_class *cl = (struct htb_class *)arg;
1244 struct htb_sched *q = qdisc_priv(sch) <<
1245 struct nlattr *nest; 1098 struct nlattr *nest;
1246 struct tc_htb_opt opt; 1099 struct tc_htb_opt opt;
1247 1100
1248 /* Its safe to not acquire qdisc lock 1101 /* Its safe to not acquire qdisc lock. As we hold RTNL,
1249 * no change can happen on the class 1102 * no change can happen on the class parameters.
1250 */ 1103 */
1251 tcm->tcm_parent = cl->parent ? cl->pa 1104 tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT;
1252 tcm->tcm_handle = cl->common.classid; 1105 tcm->tcm_handle = cl->common.classid;
1253 if (!cl->level && cl->leaf.q) !! 1106 if (!cl->level && cl->un.leaf.q)
1254 tcm->tcm_info = cl->leaf.q->h !! 1107 tcm->tcm_info = cl->un.leaf.q->handle;
1255 1108
1256 nest = nla_nest_start_noflag(skb, TCA !! 1109 nest = nla_nest_start(skb, TCA_OPTIONS);
1257 if (nest == NULL) 1110 if (nest == NULL)
1258 goto nla_put_failure; 1111 goto nla_put_failure;
1259 1112
1260 memset(&opt, 0, sizeof(opt)); 1113 memset(&opt, 0, sizeof(opt));
1261 1114
1262 psched_ratecfg_getrate(&opt.rate, &cl 1115 psched_ratecfg_getrate(&opt.rate, &cl->rate);
1263 opt.buffer = PSCHED_NS2TICKS(cl->buff 1116 opt.buffer = PSCHED_NS2TICKS(cl->buffer);
1264 psched_ratecfg_getrate(&opt.ceil, &cl 1117 psched_ratecfg_getrate(&opt.ceil, &cl->ceil);
1265 opt.cbuffer = PSCHED_NS2TICKS(cl->cbu 1118 opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer);
1266 opt.quantum = cl->quantum; 1119 opt.quantum = cl->quantum;
1267 opt.prio = cl->prio; 1120 opt.prio = cl->prio;
1268 opt.level = cl->level; 1121 opt.level = cl->level;
1269 if (nla_put(skb, TCA_HTB_PARMS, sizeo 1122 if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt))
1270 goto nla_put_failure; 1123 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 1124 if ((cl->rate.rate_bytes_ps >= (1ULL << 32)) &&
1274 nla_put_u64_64bit(skb, TCA_HTB_RA 1125 nla_put_u64_64bit(skb, TCA_HTB_RATE64, cl->rate.rate_bytes_ps,
1275 TCA_HTB_PAD)) 1126 TCA_HTB_PAD))
1276 goto nla_put_failure; 1127 goto nla_put_failure;
1277 if ((cl->ceil.rate_bytes_ps >= (1ULL 1128 if ((cl->ceil.rate_bytes_ps >= (1ULL << 32)) &&
1278 nla_put_u64_64bit(skb, TCA_HTB_CE 1129 nla_put_u64_64bit(skb, TCA_HTB_CEIL64, cl->ceil.rate_bytes_ps,
1279 TCA_HTB_PAD)) 1130 TCA_HTB_PAD))
1280 goto nla_put_failure; 1131 goto nla_put_failure;
1281 1132
1282 return nla_nest_end(skb, nest); 1133 return nla_nest_end(skb, nest);
1283 1134
1284 nla_put_failure: 1135 nla_put_failure:
1285 nla_nest_cancel(skb, nest); 1136 nla_nest_cancel(skb, nest);
1286 return -1; 1137 return -1;
1287 } 1138 }
1288 1139
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 1140 static int
1320 htb_dump_class_stats(struct Qdisc *sch, unsig 1141 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d)
1321 { 1142 {
1322 struct htb_class *cl = (struct htb_cl 1143 struct htb_class *cl = (struct htb_class *)arg;
1323 struct htb_sched *q = qdisc_priv(sch) <<
1324 struct gnet_stats_queue qs = { 1144 struct gnet_stats_queue qs = {
1325 .drops = cl->drops, 1145 .drops = cl->drops,
1326 .overlimits = cl->overlimits, <<
1327 }; 1146 };
1328 __u32 qlen = 0; 1147 __u32 qlen = 0;
1329 1148
1330 if (!cl->level && cl->leaf.q) !! 1149 if (!cl->level && cl->un.leaf.q) {
1331 qdisc_qstats_qlen_backlog(cl- !! 1150 qlen = cl->un.leaf.q->q.qlen;
1332 !! 1151 qs.backlog = cl->un.leaf.q->qstats.backlog;
>> 1152 }
1333 cl->xstats.tokens = clamp_t(s64, PSCH 1153 cl->xstats.tokens = clamp_t(s64, PSCHED_NS2TICKS(cl->tokens),
1334 INT_MIN, 1154 INT_MIN, INT_MAX);
1335 cl->xstats.ctokens = clamp_t(s64, PSC 1155 cl->xstats.ctokens = clamp_t(s64, PSCHED_NS2TICKS(cl->ctokens),
1336 INT_MIN, 1156 INT_MIN, INT_MAX);
1337 1157
1338 if (q->offload) { !! 1158 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
1339 if (!cl->level) { !! 1159 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-> 1160 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1354 gnet_stats_copy_queue(d, NULL, &q 1161 gnet_stats_copy_queue(d, NULL, &qs, qlen) < 0)
1355 return -1; 1162 return -1;
1356 1163
1357 return gnet_stats_copy_app(d, &cl->xs 1164 return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1358 } 1165 }
1359 1166
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 1167 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1442 struct Qdisc **old, stru !! 1168 struct Qdisc **old)
1443 { 1169 {
1444 struct netdev_queue *dev_queue = sch- <<
1445 struct htb_class *cl = (struct htb_cl 1170 struct htb_class *cl = (struct htb_class *)arg;
1446 struct htb_sched *q = qdisc_priv(sch) <<
1447 struct Qdisc *old_q; <<
1448 1171
1449 if (cl->level) 1172 if (cl->level)
1450 return -EINVAL; 1173 return -EINVAL;
>> 1174 if (new == NULL &&
>> 1175 (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
>> 1176 cl->common.classid)) == NULL)
>> 1177 return -ENOBUFS;
1451 1178
1452 if (q->offload) !! 1179 *old = qdisc_replace(sch, new, &cl->un.leaf.q);
1453 dev_queue = htb_offload_get_q <<
1454 <<
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 <<
1468 *old = qdisc_replace(sch, new, &cl->l <<
1469 <<
1470 if (q->offload) { <<
1471 WARN_ON(old_q != *old); <<
1472 qdisc_put(old_q); <<
1473 } <<
1474 <<
1475 return 0; 1180 return 0;
1476 } 1181 }
1477 1182
1478 static struct Qdisc *htb_leaf(struct Qdisc *s 1183 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg)
1479 { 1184 {
1480 struct htb_class *cl = (struct htb_cl 1185 struct htb_class *cl = (struct htb_class *)arg;
1481 return !cl->level ? cl->leaf.q : NULL !! 1186 return !cl->level ? cl->un.leaf.q : NULL;
1482 } 1187 }
1483 1188
1484 static void htb_qlen_notify(struct Qdisc *sch 1189 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg)
1485 { 1190 {
1486 struct htb_class *cl = (struct htb_cl 1191 struct htb_class *cl = (struct htb_class *)arg;
1487 1192
1488 htb_deactivate(qdisc_priv(sch), cl); 1193 htb_deactivate(qdisc_priv(sch), cl);
1489 } 1194 }
1490 1195
1491 static inline int htb_parent_last_child(struc 1196 static inline int htb_parent_last_child(struct htb_class *cl)
1492 { 1197 {
1493 if (!cl->parent) 1198 if (!cl->parent)
1494 /* the root class */ 1199 /* the root class */
1495 return 0; 1200 return 0;
1496 if (cl->parent->children > 1) 1201 if (cl->parent->children > 1)
1497 /* not the last child */ 1202 /* not the last child */
1498 return 0; 1203 return 0;
1499 return 1; 1204 return 1;
1500 } 1205 }
1501 1206
1502 static void htb_parent_to_leaf(struct Qdisc * !! 1207 static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl,
1503 struct Qdisc * 1208 struct Qdisc *new_q)
1504 { 1209 {
1505 struct htb_sched *q = qdisc_priv(sch) <<
1506 struct htb_class *parent = cl->parent 1210 struct htb_class *parent = cl->parent;
1507 1211
1508 WARN_ON(cl->level || !cl->leaf.q || c !! 1212 WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity);
1509 1213
1510 if (parent->cmode != HTB_CAN_SEND) 1214 if (parent->cmode != HTB_CAN_SEND)
1511 htb_safe_rb_erase(&parent->pq 1215 htb_safe_rb_erase(&parent->pq_node,
1512 &q->hlevel[ 1216 &q->hlevel[parent->level].wait_pq);
1513 1217
1514 parent->level = 0; 1218 parent->level = 0;
1515 memset(&parent->inner, 0, sizeof(pare !! 1219 memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1516 parent->leaf.q = new_q ? new_q : &noo !! 1220 INIT_LIST_HEAD(&parent->un.leaf.drop_list);
>> 1221 parent->un.leaf.q = new_q ? new_q : &noop_qdisc;
1517 parent->tokens = parent->buffer; 1222 parent->tokens = parent->buffer;
1518 parent->ctokens = parent->cbuffer; 1223 parent->ctokens = parent->cbuffer;
1519 parent->t_c = ktime_get_ns(); 1224 parent->t_c = ktime_get_ns();
1520 parent->cmode = HTB_CAN_SEND; 1225 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 } 1226 }
1600 1227
1601 static void htb_destroy_class(struct Qdisc *s 1228 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
1602 { 1229 {
1603 if (!cl->level) { 1230 if (!cl->level) {
1604 WARN_ON(!cl->leaf.q); !! 1231 WARN_ON(!cl->un.leaf.q);
1605 qdisc_put(cl->leaf.q); !! 1232 qdisc_destroy(cl->un.leaf.q);
1606 } 1233 }
1607 gen_kill_estimator(&cl->rate_est); 1234 gen_kill_estimator(&cl->rate_est);
1608 tcf_block_put(cl->block); 1235 tcf_block_put(cl->block);
1609 kfree(cl); 1236 kfree(cl);
1610 } 1237 }
1611 1238
1612 static void htb_destroy(struct Qdisc *sch) 1239 static void htb_destroy(struct Qdisc *sch)
1613 { 1240 {
1614 struct net_device *dev = qdisc_dev(sc <<
1615 struct tc_htb_qopt_offload offload_op <<
1616 struct htb_sched *q = qdisc_priv(sch) 1241 struct htb_sched *q = qdisc_priv(sch);
1617 struct hlist_node *next; 1242 struct hlist_node *next;
1618 bool nonempty, changed; <<
1619 struct htb_class *cl; 1243 struct htb_class *cl;
1620 unsigned int i; 1244 unsigned int i;
1621 1245
1622 cancel_work_sync(&q->work); 1246 cancel_work_sync(&q->work);
1623 qdisc_watchdog_cancel(&q->watchdog); 1247 qdisc_watchdog_cancel(&q->watchdog);
1624 /* This line used to be after htb_des 1248 /* This line used to be after htb_destroy_class call below
1625 * and surprisingly it worked in 2.4. 1249 * and surprisingly it worked in 2.4. But it must precede it
1626 * because filter need its target cla 1250 * because filter need its target class alive to be able to call
1627 * unbind_filter on it (without Oops) 1251 * unbind_filter on it (without Oops).
1628 */ 1252 */
1629 tcf_block_put(q->block); 1253 tcf_block_put(q->block);
1630 1254
1631 for (i = 0; i < q->clhash.hashsize; i 1255 for (i = 0; i < q->clhash.hashsize; i++) {
1632 hlist_for_each_entry(cl, &q-> 1256 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
1633 tcf_block_put(cl->blo 1257 tcf_block_put(cl->block);
1634 cl->block = NULL; 1258 cl->block = NULL;
1635 } 1259 }
1636 } 1260 }
1637 !! 1261 for (i = 0; i < q->clhash.hashsize; i++) {
1638 do { !! 1262 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
1639 nonempty = false; !! 1263 common.hnode)
1640 changed = false; !! 1264 htb_destroy_class(sch, cl);
1641 for (i = 0; i < q->clhash.has !! 1265 }
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); 1266 qdisc_class_hash_destroy(&q->clhash);
1674 __qdisc_reset_queue(&q->direct_queue) 1267 __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 } 1268 }
1689 1269
1690 static int htb_delete(struct Qdisc *sch, unsi !! 1270 static int htb_delete(struct Qdisc *sch, unsigned long arg)
1691 struct netlink_ext_ack <<
1692 { 1271 {
1693 struct htb_sched *q = qdisc_priv(sch) 1272 struct htb_sched *q = qdisc_priv(sch);
1694 struct htb_class *cl = (struct htb_cl 1273 struct htb_class *cl = (struct htb_class *)arg;
1695 struct Qdisc *new_q = NULL; 1274 struct Qdisc *new_q = NULL;
1696 int last_child = 0; 1275 int last_child = 0;
1697 int err; <<
1698 1276
1699 /* TODO: why don't allow to delete su 1277 /* TODO: why don't allow to delete subtree ? references ? does
1700 * tc subsys guarantee us that in htb 1278 * tc subsys guarantee us that in htb_destroy it holds no class
1701 * refs so that we can remove childre 1279 * refs so that we can remove children safely there ?
1702 */ 1280 */
1703 if (cl->children || qdisc_class_in_us !! 1281 if (cl->children || cl->filter_cnt)
1704 NL_SET_ERR_MSG(extack, "HTB c <<
1705 return -EBUSY; 1282 return -EBUSY;
1706 } <<
1707 1283
1708 if (!cl->level && htb_parent_last_chi !! 1284 if (!cl->level && htb_parent_last_child(cl)) {
>> 1285 new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
>> 1286 cl->parent->common.classid);
1709 last_child = 1; 1287 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 <<
1721 if (q->offload) <<
1722 dev_queue = htb_offlo <<
1723 <<
1724 new_q = qdisc_create_dflt(dev <<
1725 cl- <<
1726 NUL <<
1727 if (q->offload) <<
1728 htb_parent_to_leaf_of <<
1729 } 1288 }
1730 1289
1731 sch_tree_lock(sch); 1290 sch_tree_lock(sch);
1732 1291
1733 if (!cl->level) !! 1292 if (!cl->level) {
1734 qdisc_purge_queue(cl->leaf.q) !! 1293 unsigned int qlen = cl->un.leaf.q->q.qlen;
>> 1294 unsigned int backlog = cl->un.leaf.q->qstats.backlog;
>> 1295
>> 1296 qdisc_reset(cl->un.leaf.q);
>> 1297 qdisc_tree_reduce_backlog(cl->un.leaf.q, qlen, backlog);
>> 1298 }
1735 1299
1736 /* delete from hash and active; remai 1300 /* delete from hash and active; remainder in destroy_class */
1737 qdisc_class_hash_remove(&q->clhash, & 1301 qdisc_class_hash_remove(&q->clhash, &cl->common);
1738 if (cl->parent) 1302 if (cl->parent)
1739 cl->parent->children--; 1303 cl->parent->children--;
1740 1304
1741 if (cl->prio_activity) 1305 if (cl->prio_activity)
1742 htb_deactivate(q, cl); 1306 htb_deactivate(q, cl);
1743 1307
1744 if (cl->cmode != HTB_CAN_SEND) 1308 if (cl->cmode != HTB_CAN_SEND)
1745 htb_safe_rb_erase(&cl->pq_nod 1309 htb_safe_rb_erase(&cl->pq_node,
1746 &q->hlevel[ 1310 &q->hlevel[cl->level].wait_pq);
1747 1311
1748 if (last_child) 1312 if (last_child)
1749 htb_parent_to_leaf(sch, cl, n !! 1313 htb_parent_to_leaf(q, cl, new_q);
1750 1314
1751 sch_tree_unlock(sch); 1315 sch_tree_unlock(sch);
1752 1316
1753 htb_destroy_class(sch, cl); 1317 htb_destroy_class(sch, cl);
1754 return 0; 1318 return 0;
1755 } 1319 }
1756 1320
1757 static int htb_change_class(struct Qdisc *sch 1321 static int htb_change_class(struct Qdisc *sch, u32 classid,
1758 u32 parentid, str 1322 u32 parentid, struct nlattr **tca,
1759 unsigned long *ar !! 1323 unsigned long *arg)
1760 { 1324 {
1761 int err = -EINVAL; 1325 int err = -EINVAL;
1762 struct htb_sched *q = qdisc_priv(sch) 1326 struct htb_sched *q = qdisc_priv(sch);
1763 struct htb_class *cl = (struct htb_cl 1327 struct htb_class *cl = (struct htb_class *)*arg, *parent;
1764 struct tc_htb_qopt_offload offload_op <<
1765 struct nlattr *opt = tca[TCA_OPTIONS] 1328 struct nlattr *opt = tca[TCA_OPTIONS];
1766 struct nlattr *tb[TCA_HTB_MAX + 1]; 1329 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; 1330 struct tc_htb_opt *hopt;
1770 u64 rate64, ceil64; 1331 u64 rate64, ceil64;
1771 int warn = 0; <<
1772 1332
1773 /* extract all subattrs from opt attr 1333 /* extract all subattrs from opt attr */
1774 if (!opt) 1334 if (!opt)
1775 goto failure; 1335 goto failure;
1776 1336
1777 err = nla_parse_nested_deprecated(tb, !! 1337 err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy, NULL);
1778 ext <<
1779 if (err < 0) 1338 if (err < 0)
1780 goto failure; 1339 goto failure;
1781 1340
1782 err = -EINVAL; 1341 err = -EINVAL;
1783 if (tb[TCA_HTB_PARMS] == NULL) 1342 if (tb[TCA_HTB_PARMS] == NULL)
1784 goto failure; 1343 goto failure;
1785 1344
1786 parent = parentid == TC_H_ROOT ? NULL 1345 parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch);
1787 1346
1788 hopt = nla_data(tb[TCA_HTB_PARMS]); 1347 hopt = nla_data(tb[TCA_HTB_PARMS]);
1789 if (!hopt->rate.rate || !hopt->ceil.r 1348 if (!hopt->rate.rate || !hopt->ceil.rate)
1790 goto failure; 1349 goto failure;
1791 1350
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 1351 /* Keeping backward compatible with rate_table based iproute2 tc */
1805 if (hopt->rate.linklayer == TC_LINKLA 1352 if (hopt->rate.linklayer == TC_LINKLAYER_UNAWARE)
1806 qdisc_put_rtab(qdisc_get_rtab !! 1353 qdisc_put_rtab(qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB]));
1807 <<
1808 1354
1809 if (hopt->ceil.linklayer == TC_LINKLA 1355 if (hopt->ceil.linklayer == TC_LINKLAYER_UNAWARE)
1810 qdisc_put_rtab(qdisc_get_rtab !! 1356 qdisc_put_rtab(qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB]));
1811 <<
1812 <<
1813 rate64 = tb[TCA_HTB_RATE64] ? nla_get <<
1814 ceil64 = tb[TCA_HTB_CEIL64] ? nla_get <<
1815 1357
1816 if (!cl) { /* new class 1358 if (!cl) { /* new class */
1817 struct net_device *dev = qdis !! 1359 struct Qdisc *new_q;
1818 struct Qdisc *new_q, *old_q; <<
1819 int prio; 1360 int prio;
1820 struct { 1361 struct {
1821 struct nlattr 1362 struct nlattr nla;
1822 struct gnet_estimator 1363 struct gnet_estimator opt;
1823 } est = { 1364 } est = {
1824 .nla = { 1365 .nla = {
1825 .nla_len 1366 .nla_len = nla_attr_size(sizeof(est.opt)),
1826 .nla_type 1367 .nla_type = TCA_RATE,
1827 }, 1368 },
1828 .opt = { 1369 .opt = {
1829 /* 4s interva 1370 /* 4s interval, 16s averaging constant */
1830 .interval 1371 .interval = 2,
1831 .ewma_log 1372 .ewma_log = 2,
1832 }, 1373 },
1833 }; 1374 };
1834 1375
1835 /* check for valid classid */ 1376 /* check for valid classid */
1836 if (!classid || TC_H_MAJ(clas 1377 if (!classid || TC_H_MAJ(classid ^ sch->handle) ||
1837 htb_find(classid, sch)) 1378 htb_find(classid, sch))
1838 goto failure; 1379 goto failure;
1839 1380
1840 /* check maximal depth */ 1381 /* check maximal depth */
1841 if (parent && parent->parent 1382 if (parent && parent->parent && parent->parent->level < 2) {
1842 NL_SET_ERR_MSG_MOD(ex !! 1383 pr_err("htb: tree is too deep\n");
1843 goto failure; 1384 goto failure;
1844 } 1385 }
1845 err = -ENOBUFS; 1386 err = -ENOBUFS;
1846 cl = kzalloc(sizeof(*cl), GFP 1387 cl = kzalloc(sizeof(*cl), GFP_KERNEL);
1847 if (!cl) 1388 if (!cl)
1848 goto failure; 1389 goto failure;
1849 1390
1850 gnet_stats_basic_sync_init(&c !! 1391 err = tcf_block_get(&cl->block, &cl->filter_list);
1851 gnet_stats_basic_sync_init(&c <<
1852 <<
1853 err = tcf_block_get(&cl->bloc <<
1854 if (err) { 1392 if (err) {
1855 kfree(cl); 1393 kfree(cl);
1856 goto failure; 1394 goto failure;
1857 } 1395 }
1858 if (htb_rate_est || tca[TCA_R 1396 if (htb_rate_est || tca[TCA_RATE]) {
1859 err = gen_new_estimat 1397 err = gen_new_estimator(&cl->bstats, NULL,
1860 1398 &cl->rate_est,
1861 1399 NULL,
1862 !! 1400 qdisc_root_sleeping_running(sch),
1863 1401 tca[TCA_RATE] ? : &est.nla);
1864 if (err) !! 1402 if (err) {
1865 goto err_bloc !! 1403 tcf_block_put(cl->block);
>> 1404 kfree(cl);
>> 1405 goto failure;
>> 1406 }
1866 } 1407 }
1867 1408
1868 cl->children = 0; 1409 cl->children = 0;
>> 1410 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
1869 RB_CLEAR_NODE(&cl->pq_node); 1411 RB_CLEAR_NODE(&cl->pq_node);
1870 1412
1871 for (prio = 0; prio < TC_HTB_ 1413 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++)
1872 RB_CLEAR_NODE(&cl->no 1414 RB_CLEAR_NODE(&cl->node[prio]);
1873 1415
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 1416 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1882 * so that can't be used insi 1417 * so that can't be used inside of sch_tree_lock
1883 * -- thanks to Karlis Peisen 1418 * -- thanks to Karlis Peisenieks
1884 */ 1419 */
1885 if (!q->offload) { !! 1420 new_q = qdisc_create_dflt(sch->dev_queue,
1886 dev_queue = sch->dev_ !! 1421 &pfifo_qdisc_ops, classid);
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 <<
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); 1422 sch_tree_lock(sch);
1946 if (parent && !parent->level) 1423 if (parent && !parent->level) {
>> 1424 unsigned int qlen = parent->un.leaf.q->q.qlen;
>> 1425 unsigned int backlog = parent->un.leaf.q->qstats.backlog;
>> 1426
1947 /* turn parent into i 1427 /* turn parent into inner node */
1948 qdisc_purge_queue(par !! 1428 qdisc_reset(parent->un.leaf.q);
1949 parent_qdisc = parent !! 1429 qdisc_tree_reduce_backlog(parent->un.leaf.q, qlen, backlog);
>> 1430 qdisc_destroy(parent->un.leaf.q);
1950 if (parent->prio_acti 1431 if (parent->prio_activity)
1951 htb_deactivat 1432 htb_deactivate(q, parent);
1952 1433
1953 /* remove from evt li 1434 /* remove from evt list because of level change */
1954 if (parent->cmode != 1435 if (parent->cmode != HTB_CAN_SEND) {
1955 htb_safe_rb_e 1436 htb_safe_rb_erase(&parent->pq_node, &q->hlevel[0].wait_pq);
1956 parent->cmode 1437 parent->cmode = HTB_CAN_SEND;
1957 } 1438 }
1958 parent->level = (pare 1439 parent->level = (parent->parent ? parent->parent->level
1959 : TC 1440 : TC_HTB_MAXDEPTH) - 1;
1960 memset(&parent->inner !! 1441 memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1961 } 1442 }
1962 <<
1963 /* leaf (we) needs elementary 1443 /* leaf (we) needs elementary qdisc */
1964 cl->leaf.q = new_q ? new_q : !! 1444 cl->un.leaf.q = new_q ? new_q : &noop_qdisc;
1965 if (q->offload) <<
1966 cl->leaf.offload_queu <<
1967 1445
>> 1446 cl->common.classid = classid;
1968 cl->parent = parent; 1447 cl->parent = parent;
1969 1448
1970 /* set class to be in HTB_CAN 1449 /* set class to be in HTB_CAN_SEND state */
1971 cl->tokens = PSCHED_TICKS2NS( 1450 cl->tokens = PSCHED_TICKS2NS(hopt->buffer);
1972 cl->ctokens = PSCHED_TICKS2NS 1451 cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer);
1973 cl->mbuffer = 60ULL * NSEC_PE 1452 cl->mbuffer = 60ULL * NSEC_PER_SEC; /* 1min */
1974 cl->t_c = ktime_get_ns(); 1453 cl->t_c = ktime_get_ns();
1975 cl->cmode = HTB_CAN_SEND; 1454 cl->cmode = HTB_CAN_SEND;
1976 1455
1977 /* attach to the hash list an 1456 /* attach to the hash list and parent's family */
1978 qdisc_class_hash_insert(&q->c 1457 qdisc_class_hash_insert(&q->clhash, &cl->common);
1979 if (parent) 1458 if (parent)
1980 parent->children++; 1459 parent->children++;
1981 if (cl->leaf.q != &noop_qdisc !! 1460 if (cl->un.leaf.q != &noop_qdisc)
1982 qdisc_hash_add(cl->le !! 1461 qdisc_hash_add(cl->un.leaf.q, true);
1983 } else { 1462 } else {
1984 if (tca[TCA_RATE]) { 1463 if (tca[TCA_RATE]) {
1985 err = gen_replace_est 1464 err = gen_replace_estimator(&cl->bstats, NULL,
1986 1465 &cl->rate_est,
1987 1466 NULL,
1988 !! 1467 qdisc_root_sleeping_running(sch),
1989 1468 tca[TCA_RATE]);
1990 if (err) 1469 if (err)
1991 return err; 1470 return err;
1992 } 1471 }
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); 1472 sch_tree_lock(sch);
2018 } 1473 }
2019 1474
>> 1475 rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0;
>> 1476
>> 1477 ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0;
>> 1478
2020 psched_ratecfg_precompute(&cl->rate, 1479 psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64);
2021 psched_ratecfg_precompute(&cl->ceil, 1480 psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64);
2022 1481
2023 /* it used to be a nasty bug here, we 1482 /* it used to be a nasty bug here, we have to check that node
2024 * is really leaf before changing cl- !! 1483 * is really leaf before changing cl->un.leaf !
2025 */ 1484 */
2026 if (!cl->level) { 1485 if (!cl->level) {
2027 u64 quantum = cl->rate.rate_b 1486 u64 quantum = cl->rate.rate_bytes_ps;
2028 1487
2029 do_div(quantum, q->rate2quant 1488 do_div(quantum, q->rate2quantum);
2030 cl->quantum = min_t(u64, quan 1489 cl->quantum = min_t(u64, quantum, INT_MAX);
2031 1490
2032 if (!hopt->quantum && cl->qua 1491 if (!hopt->quantum && cl->quantum < 1000) {
2033 warn = -1; !! 1492 pr_warn("HTB: quantum of class %X is small. Consider r2q change.\n",
>> 1493 cl->common.classid);
2034 cl->quantum = 1000; 1494 cl->quantum = 1000;
2035 } 1495 }
2036 if (!hopt->quantum && cl->qua 1496 if (!hopt->quantum && cl->quantum > 200000) {
2037 warn = 1; !! 1497 pr_warn("HTB: quantum of class %X is big. Consider r2q change.\n",
>> 1498 cl->common.classid);
2038 cl->quantum = 200000; 1499 cl->quantum = 200000;
2039 } 1500 }
2040 if (hopt->quantum) 1501 if (hopt->quantum)
2041 cl->quantum = hopt->q 1502 cl->quantum = hopt->quantum;
2042 if ((cl->prio = hopt->prio) > 1503 if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO)
2043 cl->prio = TC_HTB_NUM 1504 cl->prio = TC_HTB_NUMPRIO - 1;
2044 } 1505 }
2045 1506
2046 cl->buffer = PSCHED_TICKS2NS(hopt->bu 1507 cl->buffer = PSCHED_TICKS2NS(hopt->buffer);
2047 cl->cbuffer = PSCHED_TICKS2NS(hopt->c 1508 cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer);
2048 1509
2049 sch_tree_unlock(sch); 1510 sch_tree_unlock(sch);
2050 qdisc_put(parent_qdisc); <<
2051 <<
2052 if (warn) <<
2053 NL_SET_ERR_MSG_FMT_MOD(extack <<
2054 "quant <<
2055 cl->co <<
2056 1511
2057 qdisc_class_hash_grow(sch, &q->clhash 1512 qdisc_class_hash_grow(sch, &q->clhash);
2058 1513
2059 *arg = (unsigned long)cl; 1514 *arg = (unsigned long)cl;
2060 return 0; 1515 return 0;
2061 1516
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: 1517 failure:
2068 return err; 1518 return err;
2069 } 1519 }
2070 1520
2071 static struct tcf_block *htb_tcf_block(struct !! 1521 static struct tcf_block *htb_tcf_block(struct Qdisc *sch, unsigned long arg)
2072 struct <<
2073 { 1522 {
2074 struct htb_sched *q = qdisc_priv(sch) 1523 struct htb_sched *q = qdisc_priv(sch);
2075 struct htb_class *cl = (struct htb_cl 1524 struct htb_class *cl = (struct htb_class *)arg;
2076 1525
2077 return cl ? cl->block : q->block; 1526 return cl ? cl->block : q->block;
2078 } 1527 }
2079 1528
2080 static unsigned long htb_bind_filter(struct Q 1529 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
2081 u32 clas 1530 u32 classid)
2082 { 1531 {
2083 struct htb_class *cl = htb_find(class 1532 struct htb_class *cl = htb_find(classid, sch);
2084 1533
2085 /*if (cl && !cl->level) return 0; 1534 /*if (cl && !cl->level) return 0;
2086 * The line above used to be there to 1535 * The line above used to be there to prevent attaching filters to
2087 * leaves. But at least tc_index filt 1536 * leaves. But at least tc_index filter uses this just to get class
2088 * for other reasons so that we have 1537 * for other reasons so that we have to allow for it.
2089 * ---- 1538 * ----
2090 * 19.6.2002 As Werner explained it i 1539 * 19.6.2002 As Werner explained it is ok - bind filter is just
2091 * another way to "lock" the class - 1540 * another way to "lock" the class - unlike "get" this lock can
2092 * be broken by class during destroy 1541 * be broken by class during destroy IIUC.
2093 */ 1542 */
2094 if (cl) 1543 if (cl)
2095 qdisc_class_get(&cl->common); !! 1544 cl->filter_cnt++;
2096 return (unsigned long)cl; 1545 return (unsigned long)cl;
2097 } 1546 }
2098 1547
2099 static void htb_unbind_filter(struct Qdisc *s 1548 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
2100 { 1549 {
2101 struct htb_class *cl = (struct htb_cl 1550 struct htb_class *cl = (struct htb_class *)arg;
2102 1551
2103 qdisc_class_put(&cl->common); !! 1552 if (cl)
>> 1553 cl->filter_cnt--;
2104 } 1554 }
2105 1555
2106 static void htb_walk(struct Qdisc *sch, struc 1556 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
2107 { 1557 {
2108 struct htb_sched *q = qdisc_priv(sch) 1558 struct htb_sched *q = qdisc_priv(sch);
2109 struct htb_class *cl; 1559 struct htb_class *cl;
2110 unsigned int i; 1560 unsigned int i;
2111 1561
2112 if (arg->stop) 1562 if (arg->stop)
2113 return; 1563 return;
2114 1564
2115 for (i = 0; i < q->clhash.hashsize; i 1565 for (i = 0; i < q->clhash.hashsize; i++) {
2116 hlist_for_each_entry(cl, &q-> 1566 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
2117 if (!tc_qdisc_stats_d !! 1567 if (arg->count < arg->skip) {
>> 1568 arg->count++;
>> 1569 continue;
>> 1570 }
>> 1571 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
>> 1572 arg->stop = 1;
2118 return; 1573 return;
>> 1574 }
>> 1575 arg->count++;
2119 } 1576 }
2120 } 1577 }
2121 } 1578 }
2122 1579
2123 static const struct Qdisc_class_ops htb_class 1580 static const struct Qdisc_class_ops htb_class_ops = {
2124 .select_queue = htb_select_qu <<
2125 .graft = htb_graft, 1581 .graft = htb_graft,
2126 .leaf = htb_leaf, 1582 .leaf = htb_leaf,
2127 .qlen_notify = htb_qlen_noti 1583 .qlen_notify = htb_qlen_notify,
2128 .find = htb_search, 1584 .find = htb_search,
2129 .change = htb_change_cl 1585 .change = htb_change_class,
2130 .delete = htb_delete, 1586 .delete = htb_delete,
2131 .walk = htb_walk, 1587 .walk = htb_walk,
2132 .tcf_block = htb_tcf_block 1588 .tcf_block = htb_tcf_block,
2133 .bind_tcf = htb_bind_filt 1589 .bind_tcf = htb_bind_filter,
2134 .unbind_tcf = htb_unbind_fi 1590 .unbind_tcf = htb_unbind_filter,
2135 .dump = htb_dump_clas 1591 .dump = htb_dump_class,
2136 .dump_stats = htb_dump_clas 1592 .dump_stats = htb_dump_class_stats,
2137 }; 1593 };
2138 1594
2139 static struct Qdisc_ops htb_qdisc_ops __read_ 1595 static struct Qdisc_ops htb_qdisc_ops __read_mostly = {
2140 .cl_ops = &htb_class_op 1596 .cl_ops = &htb_class_ops,
2141 .id = "htb", 1597 .id = "htb",
2142 .priv_size = sizeof(struct 1598 .priv_size = sizeof(struct htb_sched),
2143 .enqueue = htb_enqueue, 1599 .enqueue = htb_enqueue,
2144 .dequeue = htb_dequeue, 1600 .dequeue = htb_dequeue,
2145 .peek = qdisc_peek_de 1601 .peek = qdisc_peek_dequeued,
2146 .init = htb_init, 1602 .init = htb_init,
2147 .attach = htb_attach, <<
2148 .reset = htb_reset, 1603 .reset = htb_reset,
2149 .destroy = htb_destroy, 1604 .destroy = htb_destroy,
2150 .dump = htb_dump, 1605 .dump = htb_dump,
2151 .owner = THIS_MODULE, 1606 .owner = THIS_MODULE,
2152 }; 1607 };
2153 MODULE_ALIAS_NET_SCH("htb"); <<
2154 1608
2155 static int __init htb_module_init(void) 1609 static int __init htb_module_init(void)
2156 { 1610 {
2157 return register_qdisc(&htb_qdisc_ops) 1611 return register_qdisc(&htb_qdisc_ops);
2158 } 1612 }
2159 static void __exit htb_module_exit(void) 1613 static void __exit htb_module_exit(void)
2160 { 1614 {
2161 unregister_qdisc(&htb_qdisc_ops); 1615 unregister_qdisc(&htb_qdisc_ops);
2162 } 1616 }
2163 1617
2164 module_init(htb_module_init) 1618 module_init(htb_module_init)
2165 module_exit(htb_module_exit) 1619 module_exit(htb_module_exit)
2166 MODULE_LICENSE("GPL"); 1620 MODULE_LICENSE("GPL");
2167 MODULE_DESCRIPTION("Hierarchical Token Bucket <<
2168 1621
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.