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