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