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 supplied 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_sync bstats;
116 struct gnet_stats_basic_sync bstats_bi 117 struct gnet_stats_basic_sync bstats_bias;
117 struct tc_htb_xstats xstats; /* our 118 struct tc_htb_xstats xstats; /* our special stats */
118 119
119 /* token bucket parameters */ 120 /* token bucket parameters */
120 s64 tokens, ctoken 121 s64 tokens, ctokens;/* current number of tokens */
121 s64 t_c; 122 s64 t_c; /* checkpoint time */
122 123
123 union { 124 union {
124 struct htb_class_leaf { 125 struct htb_class_leaf {
125 int defici 126 int deficit[TC_HTB_MAXDEPTH];
126 struct Qdisc *q; 127 struct Qdisc *q;
127 struct netdev_queue *o 128 struct netdev_queue *offload_queue;
128 } leaf; 129 } leaf;
129 struct htb_class_inner { 130 struct htb_class_inner {
130 struct htb_prio clprio 131 struct htb_prio clprio[TC_HTB_NUMPRIO];
131 } inner; 132 } inner;
132 }; 133 };
133 s64 pq_key; 134 s64 pq_key;
134 135
135 int prio_activity; 136 int prio_activity; /* for which prios are we active */
136 enum htb_cmode cmode; 137 enum htb_cmode cmode; /* current mode of the class */
137 struct rb_node pq_node; 138 struct rb_node pq_node; /* node for event queue */
138 struct rb_node node[TC_HTB_NU 139 struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */
139 140
140 unsigned int drops ____cacheline_align 141 unsigned int drops ____cacheline_aligned_in_smp;
141 unsigned int overlimits; 142 unsigned int overlimits;
142 }; 143 };
143 144
144 struct htb_level { 145 struct htb_level {
145 struct rb_root wait_pq; 146 struct rb_root wait_pq;
146 struct htb_prio hprio[TC_HTB_NUMPRIO]; 147 struct htb_prio hprio[TC_HTB_NUMPRIO];
147 }; 148 };
148 149
149 struct htb_sched { 150 struct htb_sched {
150 struct Qdisc_class_hash clhash; 151 struct Qdisc_class_hash clhash;
151 int defcls; 152 int defcls; /* class where unclassified flows go to */
152 int rate2quantum; 153 int rate2quantum; /* quant = rate / rate2quantum */
153 154
154 /* filters for qdisc itself */ 155 /* filters for qdisc itself */
155 struct tcf_proto __rcu *filter_list; 156 struct tcf_proto __rcu *filter_list;
156 struct tcf_block *block; 157 struct tcf_block *block;
157 158
158 #define HTB_WARN_TOOMANYEVENTS 0x1 159 #define HTB_WARN_TOOMANYEVENTS 0x1
159 unsigned int warned; /* onl 160 unsigned int warned; /* only one warning */
160 int direct_qlen; 161 int direct_qlen;
161 struct work_struct work; 162 struct work_struct work;
162 163
163 /* non shaped skbs; let them go direct 164 /* non shaped skbs; let them go directly thru */
164 struct qdisc_skb_head direct_queue; 165 struct qdisc_skb_head direct_queue;
165 u32 direct_pkts; 166 u32 direct_pkts;
166 u32 overlimits; 167 u32 overlimits;
167 168
168 struct qdisc_watchdog watchdog; 169 struct qdisc_watchdog watchdog;
169 170
170 s64 now; /* cac 171 s64 now; /* cached dequeue time */
171 172
172 /* time of nearest event per level (ro 173 /* time of nearest event per level (row) */
173 s64 near_ev_cache[ 174 s64 near_ev_cache[TC_HTB_MAXDEPTH];
174 175
175 int row_mask[TC_HT 176 int row_mask[TC_HTB_MAXDEPTH];
176 177
177 struct htb_level hlevel[TC_HTB_ 178 struct htb_level hlevel[TC_HTB_MAXDEPTH];
178 179
179 struct Qdisc **direct_qdisc 180 struct Qdisc **direct_qdiscs;
180 unsigned int num_direct_qdi 181 unsigned int num_direct_qdiscs;
181 182
182 bool offload; 183 bool offload;
183 }; 184 };
184 185
185 /* find class in global hash table using given 186 /* find class in global hash table using given handle */
186 static inline struct htb_class *htb_find(u32 h 187 static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch)
187 { 188 {
188 struct htb_sched *q = qdisc_priv(sch); 189 struct htb_sched *q = qdisc_priv(sch);
189 struct Qdisc_class_common *clc; 190 struct Qdisc_class_common *clc;
190 191
191 clc = qdisc_class_find(&q->clhash, han 192 clc = qdisc_class_find(&q->clhash, handle);
192 if (clc == NULL) 193 if (clc == NULL)
193 return NULL; 194 return NULL;
194 return container_of(clc, struct htb_cl 195 return container_of(clc, struct htb_class, common);
195 } 196 }
196 197
197 static unsigned long htb_search(struct Qdisc * 198 static unsigned long htb_search(struct Qdisc *sch, u32 handle)
198 { 199 {
199 return (unsigned long)htb_find(handle, 200 return (unsigned long)htb_find(handle, sch);
200 } 201 }
201 <<
202 #define HTB_DIRECT ((struct htb_class *)-1L) <<
203 <<
204 /** 202 /**
205 * htb_classify - classify a packet into class 203 * 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 * 204 *
210 * It returns NULL if the packet should be dro 205 * It returns NULL if the packet should be dropped or -1 if the packet
211 * should be passed directly thru. In all othe 206 * should be passed directly thru. In all other cases leaf class is returned.
212 * We allow direct class selection by classid 207 * We allow direct class selection by classid in priority. The we examine
213 * filters in qdisc and in inner nodes (if hig 208 * 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 209 * node). If we end up with classid MAJOR:0 we enqueue the skb into special
215 * internal fifo (direct). These packets then 210 * internal fifo (direct). These packets then go directly thru. If we still
216 * have no valid leaf we try to use MAJOR:defa 211 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful
217 * then finish and return direct queue. 212 * then finish and return direct queue.
218 */ 213 */
>> 214 #define HTB_DIRECT ((struct htb_class *)-1L)
>> 215
219 static struct htb_class *htb_classify(struct s 216 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch,
220 int *qer 217 int *qerr)
221 { 218 {
222 struct htb_sched *q = qdisc_priv(sch); 219 struct htb_sched *q = qdisc_priv(sch);
223 struct htb_class *cl; 220 struct htb_class *cl;
224 struct tcf_result res; 221 struct tcf_result res;
225 struct tcf_proto *tcf; 222 struct tcf_proto *tcf;
226 int result; 223 int result;
227 224
228 /* allow to select class by setting sk 225 /* allow to select class by setting skb->priority to valid classid;
229 * note that nfmark can be used too by 226 * note that nfmark can be used too by attaching filter fw with no
230 * rules in it 227 * rules in it
231 */ 228 */
232 if (skb->priority == sch->handle) 229 if (skb->priority == sch->handle)
233 return HTB_DIRECT; /* X:0 230 return HTB_DIRECT; /* X:0 (direct flow) selected */
234 cl = htb_find(skb->priority, sch); 231 cl = htb_find(skb->priority, sch);
235 if (cl) { 232 if (cl) {
236 if (cl->level == 0) 233 if (cl->level == 0)
237 return cl; 234 return cl;
238 /* Start with inner filter cha 235 /* Start with inner filter chain if a non-leaf class is selected */
239 tcf = rcu_dereference_bh(cl->f 236 tcf = rcu_dereference_bh(cl->filter_list);
240 } else { 237 } else {
241 tcf = rcu_dereference_bh(q->fi 238 tcf = rcu_dereference_bh(q->filter_list);
242 } 239 }
243 240
244 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_ 241 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
245 while (tcf && (result = tcf_classify(s 242 while (tcf && (result = tcf_classify(skb, NULL, tcf, &res, false)) >= 0) {
246 #ifdef CONFIG_NET_CLS_ACT 243 #ifdef CONFIG_NET_CLS_ACT
247 switch (result) { 244 switch (result) {
248 case TC_ACT_QUEUED: 245 case TC_ACT_QUEUED:
249 case TC_ACT_STOLEN: 246 case TC_ACT_STOLEN:
250 case TC_ACT_TRAP: 247 case TC_ACT_TRAP:
251 *qerr = NET_XMIT_SUCCE 248 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
252 fallthrough; 249 fallthrough;
253 case TC_ACT_SHOT: 250 case TC_ACT_SHOT:
254 return NULL; 251 return NULL;
255 } 252 }
256 #endif 253 #endif
257 cl = (void *)res.class; 254 cl = (void *)res.class;
258 if (!cl) { 255 if (!cl) {
259 if (res.classid == sch 256 if (res.classid == sch->handle)
260 return HTB_DIR 257 return HTB_DIRECT; /* X:0 (direct flow) */
261 cl = htb_find(res.clas 258 cl = htb_find(res.classid, sch);
262 if (!cl) 259 if (!cl)
263 break; /* fil 260 break; /* filter selected invalid classid */
264 } 261 }
265 if (!cl->level) 262 if (!cl->level)
266 return cl; /* we 263 return cl; /* we hit leaf; return it */
267 264
268 /* we have got inner class; ap 265 /* we have got inner class; apply inner filter chain */
269 tcf = rcu_dereference_bh(cl->f 266 tcf = rcu_dereference_bh(cl->filter_list);
270 } 267 }
271 /* classification failed; try to use d 268 /* classification failed; try to use default class */
272 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch-> 269 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
273 if (!cl || cl->level) 270 if (!cl || cl->level)
274 return HTB_DIRECT; /* bad 271 return HTB_DIRECT; /* bad default .. this is safe bet */
275 return cl; 272 return cl;
276 } 273 }
277 274
278 /** 275 /**
279 * htb_add_to_id_tree - adds class to the roun 276 * htb_add_to_id_tree - adds class to the round robin list
280 * @root: the root of the tree 277 * @root: the root of the tree
281 * @cl: the class to add 278 * @cl: the class to add
282 * @prio: the give prio in class 279 * @prio: the give prio in class
283 * 280 *
284 * Routine adds class to the list (actually tr 281 * Routine adds class to the list (actually tree) sorted by classid.
285 * Make sure that class is not already on such 282 * Make sure that class is not already on such list for given prio.
286 */ 283 */
287 static void htb_add_to_id_tree(struct rb_root 284 static void htb_add_to_id_tree(struct rb_root *root,
288 struct htb_clas 285 struct htb_class *cl, int prio)
289 { 286 {
290 struct rb_node **p = &root->rb_node, * 287 struct rb_node **p = &root->rb_node, *parent = NULL;
291 288
292 while (*p) { 289 while (*p) {
293 struct htb_class *c; 290 struct htb_class *c;
294 parent = *p; 291 parent = *p;
295 c = rb_entry(parent, struct ht 292 c = rb_entry(parent, struct htb_class, node[prio]);
296 293
297 if (cl->common.classid > c->co 294 if (cl->common.classid > c->common.classid)
298 p = &parent->rb_right; 295 p = &parent->rb_right;
299 else 296 else
300 p = &parent->rb_left; 297 p = &parent->rb_left;
301 } 298 }
302 rb_link_node(&cl->node[prio], parent, 299 rb_link_node(&cl->node[prio], parent, p);
303 rb_insert_color(&cl->node[prio], root) 300 rb_insert_color(&cl->node[prio], root);
304 } 301 }
305 302
306 /** 303 /**
307 * htb_add_to_wait_tree - adds class to the ev 304 * htb_add_to_wait_tree - adds class to the event queue with delay
308 * @q: the priority event queue 305 * @q: the priority event queue
309 * @cl: the class to add 306 * @cl: the class to add
310 * @delay: delay in microseconds 307 * @delay: delay in microseconds
311 * 308 *
312 * The class is added to priority event queue 309 * The class is added to priority event queue to indicate that class will
313 * change its mode in cl->pq_key microseconds. 310 * change its mode in cl->pq_key microseconds. Make sure that class is not
314 * already in the queue. 311 * already in the queue.
315 */ 312 */
316 static void htb_add_to_wait_tree(struct htb_sc 313 static void htb_add_to_wait_tree(struct htb_sched *q,
317 struct htb_cl 314 struct htb_class *cl, s64 delay)
318 { 315 {
319 struct rb_node **p = &q->hlevel[cl->le 316 struct rb_node **p = &q->hlevel[cl->level].wait_pq.rb_node, *parent = NULL;
320 317
321 cl->pq_key = q->now + delay; 318 cl->pq_key = q->now + delay;
322 if (cl->pq_key == q->now) 319 if (cl->pq_key == q->now)
323 cl->pq_key++; 320 cl->pq_key++;
324 321
325 /* update the nearest event cache */ 322 /* update the nearest event cache */
326 if (q->near_ev_cache[cl->level] > cl-> 323 if (q->near_ev_cache[cl->level] > cl->pq_key)
327 q->near_ev_cache[cl->level] = 324 q->near_ev_cache[cl->level] = cl->pq_key;
328 325
329 while (*p) { 326 while (*p) {
330 struct htb_class *c; 327 struct htb_class *c;
331 parent = *p; 328 parent = *p;
332 c = rb_entry(parent, struct ht 329 c = rb_entry(parent, struct htb_class, pq_node);
333 if (cl->pq_key >= c->pq_key) 330 if (cl->pq_key >= c->pq_key)
334 p = &parent->rb_right; 331 p = &parent->rb_right;
335 else 332 else
336 p = &parent->rb_left; 333 p = &parent->rb_left;
337 } 334 }
338 rb_link_node(&cl->pq_node, parent, p); 335 rb_link_node(&cl->pq_node, parent, p);
339 rb_insert_color(&cl->pq_node, &q->hlev 336 rb_insert_color(&cl->pq_node, &q->hlevel[cl->level].wait_pq);
340 } 337 }
341 338
342 /** 339 /**
343 * htb_next_rb_node - finds next node in binar 340 * htb_next_rb_node - finds next node in binary tree
344 * @n: the current node in binary tree 341 * @n: the current node in binary tree
345 * 342 *
346 * When we are past last key we return NULL. 343 * When we are past last key we return NULL.
347 * Average complexity is 2 steps per call. 344 * Average complexity is 2 steps per call.
348 */ 345 */
349 static inline void htb_next_rb_node(struct rb_ 346 static inline void htb_next_rb_node(struct rb_node **n)
350 { 347 {
351 *n = rb_next(*n); 348 *n = rb_next(*n);
352 } 349 }
353 350
354 /** 351 /**
355 * htb_add_class_to_row - add class to its row 352 * htb_add_class_to_row - add class to its row
356 * @q: the priority event queue 353 * @q: the priority event queue
357 * @cl: the class to add 354 * @cl: the class to add
358 * @mask: the given priorities in class in bit 355 * @mask: the given priorities in class in bitmap
359 * 356 *
360 * The class is added to row at priorities mar 357 * The class is added to row at priorities marked in mask.
361 * It does nothing if mask == 0. 358 * It does nothing if mask == 0.
362 */ 359 */
363 static inline void htb_add_class_to_row(struct 360 static inline void htb_add_class_to_row(struct htb_sched *q,
364 struct 361 struct htb_class *cl, int mask)
365 { 362 {
366 q->row_mask[cl->level] |= mask; 363 q->row_mask[cl->level] |= mask;
367 while (mask) { 364 while (mask) {
368 int prio = ffz(~mask); 365 int prio = ffz(~mask);
369 mask &= ~(1 << prio); 366 mask &= ~(1 << prio);
370 htb_add_to_id_tree(&q->hlevel[ 367 htb_add_to_id_tree(&q->hlevel[cl->level].hprio[prio].row, cl, prio);
371 } 368 }
372 } 369 }
373 370
374 /* If this triggers, it is a bug in this code, 371 /* 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 * 372 static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root)
376 { 373 {
377 if (RB_EMPTY_NODE(rb)) { 374 if (RB_EMPTY_NODE(rb)) {
378 WARN_ON(1); 375 WARN_ON(1);
379 } else { 376 } else {
380 rb_erase(rb, root); 377 rb_erase(rb, root);
381 RB_CLEAR_NODE(rb); 378 RB_CLEAR_NODE(rb);
382 } 379 }
383 } 380 }
384 381
385 382
386 /** 383 /**
387 * htb_remove_class_from_row - removes class f 384 * htb_remove_class_from_row - removes class from its row
388 * @q: the priority event queue 385 * @q: the priority event queue
389 * @cl: the class to add 386 * @cl: the class to add
390 * @mask: the given priorities in class in bit 387 * @mask: the given priorities in class in bitmap
391 * 388 *
392 * The class is removed from row at priorities 389 * The class is removed from row at priorities marked in mask.
393 * It does nothing if mask == 0. 390 * It does nothing if mask == 0.
394 */ 391 */
395 static inline void htb_remove_class_from_row(s 392 static inline void htb_remove_class_from_row(struct htb_sched *q,
396 393 struct htb_class *cl, int mask)
397 { 394 {
398 int m = 0; 395 int m = 0;
399 struct htb_level *hlevel = &q->hlevel[ 396 struct htb_level *hlevel = &q->hlevel[cl->level];
400 397
401 while (mask) { 398 while (mask) {
402 int prio = ffz(~mask); 399 int prio = ffz(~mask);
403 struct htb_prio *hprio = &hlev 400 struct htb_prio *hprio = &hlevel->hprio[prio];
404 401
405 mask &= ~(1 << prio); 402 mask &= ~(1 << prio);
406 if (hprio->ptr == cl->node + p 403 if (hprio->ptr == cl->node + prio)
407 htb_next_rb_node(&hpri 404 htb_next_rb_node(&hprio->ptr);
408 405
409 htb_safe_rb_erase(cl->node + p 406 htb_safe_rb_erase(cl->node + prio, &hprio->row);
410 if (!hprio->row.rb_node) 407 if (!hprio->row.rb_node)
411 m |= 1 << prio; 408 m |= 1 << prio;
412 } 409 }
413 q->row_mask[cl->level] &= ~m; 410 q->row_mask[cl->level] &= ~m;
414 } 411 }
415 412
416 /** 413 /**
417 * htb_activate_prios - creates active classe' 414 * htb_activate_prios - creates active classe's feed chain
418 * @q: the priority event queue 415 * @q: the priority event queue
419 * @cl: the class to activate 416 * @cl: the class to activate
420 * 417 *
421 * The class is connected to ancestors and/or 418 * The class is connected to ancestors and/or appropriate rows
422 * for priorities it is participating on. cl-> 419 * for priorities it is participating on. cl->cmode must be new
423 * (activated) mode. It does nothing if cl->pr 420 * (activated) mode. It does nothing if cl->prio_activity == 0.
424 */ 421 */
425 static void htb_activate_prios(struct htb_sche 422 static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl)
426 { 423 {
427 struct htb_class *p = cl->parent; 424 struct htb_class *p = cl->parent;
428 long m, mask = cl->prio_activity; 425 long m, mask = cl->prio_activity;
429 426
430 while (cl->cmode == HTB_MAY_BORROW && 427 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
431 m = mask; 428 m = mask;
432 while (m) { 429 while (m) {
433 unsigned int prio = ff !! 430 int prio = ffz(~m);
434 <<
435 if (WARN_ON_ONCE(prio <<
436 break; <<
437 m &= ~(1 << prio); 431 m &= ~(1 << prio);
438 432
439 if (p->inner.clprio[pr 433 if (p->inner.clprio[prio].feed.rb_node)
440 /* parent alre 434 /* parent already has its feed in use so that
441 * reset bit i 435 * reset bit in mask as parent is already ok
442 */ 436 */
443 mask &= ~(1 << 437 mask &= ~(1 << prio);
444 438
445 htb_add_to_id_tree(&p- 439 htb_add_to_id_tree(&p->inner.clprio[prio].feed, cl, prio);
446 } 440 }
447 p->prio_activity |= mask; 441 p->prio_activity |= mask;
448 cl = p; 442 cl = p;
449 p = cl->parent; 443 p = cl->parent;
450 444
451 } 445 }
452 if (cl->cmode == HTB_CAN_SEND && mask) 446 if (cl->cmode == HTB_CAN_SEND && mask)
453 htb_add_class_to_row(q, cl, ma 447 htb_add_class_to_row(q, cl, mask);
454 } 448 }
455 449
456 /** 450 /**
457 * htb_deactivate_prios - remove class from fe 451 * htb_deactivate_prios - remove class from feed chain
458 * @q: the priority event queue 452 * @q: the priority event queue
459 * @cl: the class to deactivate 453 * @cl: the class to deactivate
460 * 454 *
461 * cl->cmode must represent old mode (before d 455 * cl->cmode must represent old mode (before deactivation). It does
462 * nothing if cl->prio_activity == 0. Class is 456 * nothing if cl->prio_activity == 0. Class is removed from all feed
463 * chains and rows. 457 * chains and rows.
464 */ 458 */
465 static void htb_deactivate_prios(struct htb_sc 459 static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl)
466 { 460 {
467 struct htb_class *p = cl->parent; 461 struct htb_class *p = cl->parent;
468 long m, mask = cl->prio_activity; 462 long m, mask = cl->prio_activity;
469 463
470 while (cl->cmode == HTB_MAY_BORROW && 464 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
471 m = mask; 465 m = mask;
472 mask = 0; 466 mask = 0;
473 while (m) { 467 while (m) {
474 int prio = ffz(~m); 468 int prio = ffz(~m);
475 m &= ~(1 << prio); 469 m &= ~(1 << prio);
476 470
477 if (p->inner.clprio[pr 471 if (p->inner.clprio[prio].ptr == cl->node + prio) {
478 /* we are remo 472 /* we are removing child which is pointed to from
479 * parent feed 473 * parent feed - forget the pointer but remember
480 * classid 474 * classid
481 */ 475 */
482 p->inner.clpri 476 p->inner.clprio[prio].last_ptr_id = cl->common.classid;
483 p->inner.clpri 477 p->inner.clprio[prio].ptr = NULL;
484 } 478 }
485 479
486 htb_safe_rb_erase(cl-> 480 htb_safe_rb_erase(cl->node + prio,
487 &p-> 481 &p->inner.clprio[prio].feed);
488 482
489 if (!p->inner.clprio[p 483 if (!p->inner.clprio[prio].feed.rb_node)
490 mask |= 1 << p 484 mask |= 1 << prio;
491 } 485 }
492 486
493 p->prio_activity &= ~mask; 487 p->prio_activity &= ~mask;
494 cl = p; 488 cl = p;
495 p = cl->parent; 489 p = cl->parent;
496 490
497 } 491 }
498 if (cl->cmode == HTB_CAN_SEND && mask) 492 if (cl->cmode == HTB_CAN_SEND && mask)
499 htb_remove_class_from_row(q, c 493 htb_remove_class_from_row(q, cl, mask);
500 } 494 }
501 495
502 static inline s64 htb_lowater(const struct htb 496 static inline s64 htb_lowater(const struct htb_class *cl)
503 { 497 {
504 if (htb_hysteresis) 498 if (htb_hysteresis)
505 return cl->cmode != HTB_CANT_S 499 return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0;
506 else 500 else
507 return 0; 501 return 0;
508 } 502 }
509 static inline s64 htb_hiwater(const struct htb 503 static inline s64 htb_hiwater(const struct htb_class *cl)
510 { 504 {
511 if (htb_hysteresis) 505 if (htb_hysteresis)
512 return cl->cmode == HTB_CAN_SE 506 return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0;
513 else 507 else
514 return 0; 508 return 0;
515 } 509 }
516 510
517 511
518 /** 512 /**
519 * htb_class_mode - computes and returns curre 513 * htb_class_mode - computes and returns current class mode
520 * @cl: the target class 514 * @cl: the target class
521 * @diff: diff time in microseconds 515 * @diff: diff time in microseconds
522 * 516 *
523 * It computes cl's mode at time cl->t_c+diff 517 * 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 518 * 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 519 * from now to time when cl will change its state.
526 * Also it is worth to note that class mode do 520 * Also it is worth to note that class mode doesn't change simply
527 * at cl->{c,}tokens == 0 but there can rather 521 * at cl->{c,}tokens == 0 but there can rather be hysteresis of
528 * 0 .. -cl->{c,}buffer range. It is meant to 522 * 0 .. -cl->{c,}buffer range. It is meant to limit number of
529 * mode transitions per time unit. The speed g 523 * mode transitions per time unit. The speed gain is about 1/6.
530 */ 524 */
531 static inline enum htb_cmode 525 static inline enum htb_cmode
532 htb_class_mode(struct htb_class *cl, s64 *diff 526 htb_class_mode(struct htb_class *cl, s64 *diff)
533 { 527 {
534 s64 toks; 528 s64 toks;
535 529
536 if ((toks = (cl->ctokens + *diff)) < h 530 if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) {
537 *diff = -toks; 531 *diff = -toks;
538 return HTB_CANT_SEND; 532 return HTB_CANT_SEND;
539 } 533 }
540 534
541 if ((toks = (cl->tokens + *diff)) >= h 535 if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl))
542 return HTB_CAN_SEND; 536 return HTB_CAN_SEND;
543 537
544 *diff = -toks; 538 *diff = -toks;
545 return HTB_MAY_BORROW; 539 return HTB_MAY_BORROW;
546 } 540 }
547 541
548 /** 542 /**
549 * htb_change_class_mode - changes classe's mo 543 * htb_change_class_mode - changes classe's mode
550 * @q: the priority event queue 544 * @q: the priority event queue
551 * @cl: the target class 545 * @cl: the target class
552 * @diff: diff time in microseconds 546 * @diff: diff time in microseconds
553 * 547 *
554 * This should be the only way how to change c 548 * This should be the only way how to change classe's mode under normal
555 * circumstances. Routine will update feed lis 549 * circumstances. Routine will update feed lists linkage, change mode
556 * and add class to the wait event queue if ap 550 * and add class to the wait event queue if appropriate. New mode should
557 * be different from old one and cl->pq_key ha 551 * 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 552 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
559 */ 553 */
560 static void 554 static void
561 htb_change_class_mode(struct htb_sched *q, str 555 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, s64 *diff)
562 { 556 {
563 enum htb_cmode new_mode = htb_class_mo 557 enum htb_cmode new_mode = htb_class_mode(cl, diff);
564 558
565 if (new_mode == cl->cmode) 559 if (new_mode == cl->cmode)
566 return; 560 return;
567 561
568 if (new_mode == HTB_CANT_SEND) { 562 if (new_mode == HTB_CANT_SEND) {
569 cl->overlimits++; 563 cl->overlimits++;
570 q->overlimits++; 564 q->overlimits++;
571 } 565 }
572 566
573 if (cl->prio_activity) { /* not 567 if (cl->prio_activity) { /* not necessary: speed optimization */
574 if (cl->cmode != HTB_CANT_SEND 568 if (cl->cmode != HTB_CANT_SEND)
575 htb_deactivate_prios(q 569 htb_deactivate_prios(q, cl);
576 cl->cmode = new_mode; 570 cl->cmode = new_mode;
577 if (new_mode != HTB_CANT_SEND) 571 if (new_mode != HTB_CANT_SEND)
578 htb_activate_prios(q, 572 htb_activate_prios(q, cl);
579 } else 573 } else
580 cl->cmode = new_mode; 574 cl->cmode = new_mode;
581 } 575 }
582 576
583 /** 577 /**
584 * htb_activate - inserts leaf cl into appropr 578 * htb_activate - inserts leaf cl into appropriate active feeds
585 * @q: the priority event queue 579 * @q: the priority event queue
586 * @cl: the target class 580 * @cl: the target class
587 * 581 *
588 * Routine learns (new) priority of leaf and a 582 * Routine learns (new) priority of leaf and activates feed chain
589 * for the prio. It can be called on already a 583 * for the prio. It can be called on already active leaf safely.
590 * It also adds leaf into droplist. 584 * It also adds leaf into droplist.
591 */ 585 */
592 static inline void htb_activate(struct htb_sch 586 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl)
593 { 587 {
594 WARN_ON(cl->level || !cl->leaf.q || !c 588 WARN_ON(cl->level || !cl->leaf.q || !cl->leaf.q->q.qlen);
595 589
596 if (!cl->prio_activity) { 590 if (!cl->prio_activity) {
597 cl->prio_activity = 1 << cl->p 591 cl->prio_activity = 1 << cl->prio;
598 htb_activate_prios(q, cl); 592 htb_activate_prios(q, cl);
599 } 593 }
600 } 594 }
601 595
602 /** 596 /**
603 * htb_deactivate - remove leaf cl from active 597 * htb_deactivate - remove leaf cl from active feeds
604 * @q: the priority event queue 598 * @q: the priority event queue
605 * @cl: the target class 599 * @cl: the target class
606 * 600 *
607 * Make sure that leaf is active. In the other 601 * Make sure that leaf is active. In the other words it can't be called
608 * with non-active leaf. It also removes class 602 * with non-active leaf. It also removes class from the drop list.
609 */ 603 */
610 static inline void htb_deactivate(struct htb_s 604 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl)
611 { 605 {
612 WARN_ON(!cl->prio_activity); 606 WARN_ON(!cl->prio_activity);
613 607
614 htb_deactivate_prios(q, cl); 608 htb_deactivate_prios(q, cl);
615 cl->prio_activity = 0; 609 cl->prio_activity = 0;
616 } 610 }
617 611
618 static int htb_enqueue(struct sk_buff *skb, st 612 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch,
619 struct sk_buff **to_fre 613 struct sk_buff **to_free)
620 { 614 {
621 int ret; 615 int ret;
622 unsigned int len = qdisc_pkt_len(skb); 616 unsigned int len = qdisc_pkt_len(skb);
623 struct htb_sched *q = qdisc_priv(sch); 617 struct htb_sched *q = qdisc_priv(sch);
624 struct htb_class *cl = htb_classify(sk 618 struct htb_class *cl = htb_classify(skb, sch, &ret);
625 619
626 if (cl == HTB_DIRECT) { 620 if (cl == HTB_DIRECT) {
627 /* enqueue to helper queue */ 621 /* enqueue to helper queue */
628 if (q->direct_queue.qlen < q-> 622 if (q->direct_queue.qlen < q->direct_qlen) {
629 __qdisc_enqueue_tail(s 623 __qdisc_enqueue_tail(skb, &q->direct_queue);
630 q->direct_pkts++; 624 q->direct_pkts++;
631 } else { 625 } else {
632 return qdisc_drop(skb, 626 return qdisc_drop(skb, sch, to_free);
633 } 627 }
634 #ifdef CONFIG_NET_CLS_ACT 628 #ifdef CONFIG_NET_CLS_ACT
635 } else if (!cl) { 629 } else if (!cl) {
636 if (ret & __NET_XMIT_BYPASS) 630 if (ret & __NET_XMIT_BYPASS)
637 qdisc_qstats_drop(sch) 631 qdisc_qstats_drop(sch);
638 __qdisc_drop(skb, to_free); 632 __qdisc_drop(skb, to_free);
639 return ret; 633 return ret;
640 #endif 634 #endif
641 } else if ((ret = qdisc_enqueue(skb, c 635 } else if ((ret = qdisc_enqueue(skb, cl->leaf.q,
642 to_fre 636 to_free)) != NET_XMIT_SUCCESS) {
643 if (net_xmit_drop_count(ret)) 637 if (net_xmit_drop_count(ret)) {
644 qdisc_qstats_drop(sch) 638 qdisc_qstats_drop(sch);
645 cl->drops++; 639 cl->drops++;
646 } 640 }
647 return ret; 641 return ret;
648 } else { 642 } else {
649 htb_activate(q, cl); 643 htb_activate(q, cl);
650 } 644 }
651 645
652 sch->qstats.backlog += len; 646 sch->qstats.backlog += len;
653 sch->q.qlen++; 647 sch->q.qlen++;
654 return NET_XMIT_SUCCESS; 648 return NET_XMIT_SUCCESS;
655 } 649 }
656 650
657 static inline void htb_accnt_tokens(struct htb 651 static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, s64 diff)
658 { 652 {
659 s64 toks = diff + cl->tokens; 653 s64 toks = diff + cl->tokens;
660 654
661 if (toks > cl->buffer) 655 if (toks > cl->buffer)
662 toks = cl->buffer; 656 toks = cl->buffer;
663 toks -= (s64) psched_l2t_ns(&cl->rate, 657 toks -= (s64) psched_l2t_ns(&cl->rate, bytes);
664 if (toks <= -cl->mbuffer) 658 if (toks <= -cl->mbuffer)
665 toks = 1 - cl->mbuffer; 659 toks = 1 - cl->mbuffer;
666 660
667 cl->tokens = toks; 661 cl->tokens = toks;
668 } 662 }
669 663
670 static inline void htb_accnt_ctokens(struct ht 664 static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, s64 diff)
671 { 665 {
672 s64 toks = diff + cl->ctokens; 666 s64 toks = diff + cl->ctokens;
673 667
674 if (toks > cl->cbuffer) 668 if (toks > cl->cbuffer)
675 toks = cl->cbuffer; 669 toks = cl->cbuffer;
676 toks -= (s64) psched_l2t_ns(&cl->ceil, 670 toks -= (s64) psched_l2t_ns(&cl->ceil, bytes);
677 if (toks <= -cl->mbuffer) 671 if (toks <= -cl->mbuffer)
678 toks = 1 - cl->mbuffer; 672 toks = 1 - cl->mbuffer;
679 673
680 cl->ctokens = toks; 674 cl->ctokens = toks;
681 } 675 }
682 676
683 /** 677 /**
684 * htb_charge_class - charges amount "bytes" t 678 * htb_charge_class - charges amount "bytes" to leaf and ancestors
685 * @q: the priority event queue 679 * @q: the priority event queue
686 * @cl: the class to start iterate 680 * @cl: the class to start iterate
687 * @level: the minimum level to account 681 * @level: the minimum level to account
688 * @skb: the socket buffer 682 * @skb: the socket buffer
689 * 683 *
690 * Routine assumes that packet "bytes" long wa 684 * Routine assumes that packet "bytes" long was dequeued from leaf cl
691 * borrowing from "level". It accounts bytes t 685 * borrowing from "level". It accounts bytes to ceil leaky bucket for
692 * leaf and all ancestors and to rate bucket f 686 * leaf and all ancestors and to rate bucket for ancestors at levels
693 * "level" and higher. It also handles possibl 687 * "level" and higher. It also handles possible change of mode resulting
694 * from the update. Note that mode can also in 688 * from the update. Note that mode can also increase here (MAY_BORROW to
695 * CAN_SEND) because we can use more precise c 689 * CAN_SEND) because we can use more precise clock that event queue here.
696 * In such case we remove class from event que 690 * In such case we remove class from event queue first.
697 */ 691 */
698 static void htb_charge_class(struct htb_sched 692 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl,
699 int level, struct 693 int level, struct sk_buff *skb)
700 { 694 {
701 int bytes = qdisc_pkt_len(skb); 695 int bytes = qdisc_pkt_len(skb);
702 enum htb_cmode old_mode; 696 enum htb_cmode old_mode;
703 s64 diff; 697 s64 diff;
704 698
705 while (cl) { 699 while (cl) {
706 diff = min_t(s64, q->now - cl- 700 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer);
707 if (cl->level >= level) { 701 if (cl->level >= level) {
708 if (cl->level == level 702 if (cl->level == level)
709 cl->xstats.len 703 cl->xstats.lends++;
710 htb_accnt_tokens(cl, b 704 htb_accnt_tokens(cl, bytes, diff);
711 } else { 705 } else {
712 cl->xstats.borrows++; 706 cl->xstats.borrows++;
713 cl->tokens += diff; 707 cl->tokens += diff; /* we moved t_c; update tokens */
714 } 708 }
715 htb_accnt_ctokens(cl, bytes, d 709 htb_accnt_ctokens(cl, bytes, diff);
716 cl->t_c = q->now; 710 cl->t_c = q->now;
717 711
718 old_mode = cl->cmode; 712 old_mode = cl->cmode;
719 diff = 0; 713 diff = 0;
720 htb_change_class_mode(q, cl, & 714 htb_change_class_mode(q, cl, &diff);
721 if (old_mode != cl->cmode) { 715 if (old_mode != cl->cmode) {
722 if (old_mode != HTB_CA 716 if (old_mode != HTB_CAN_SEND)
723 htb_safe_rb_er 717 htb_safe_rb_erase(&cl->pq_node, &q->hlevel[cl->level].wait_pq);
724 if (cl->cmode != HTB_C 718 if (cl->cmode != HTB_CAN_SEND)
725 htb_add_to_wai 719 htb_add_to_wait_tree(q, cl, diff);
726 } 720 }
727 721
728 /* update basic stats except f 722 /* update basic stats except for leaves which are already updated */
729 if (cl->level) 723 if (cl->level)
730 bstats_update(&cl->bst 724 bstats_update(&cl->bstats, skb);
731 725
732 cl = cl->parent; 726 cl = cl->parent;
733 } 727 }
734 } 728 }
735 729
736 /** 730 /**
737 * htb_do_events - make mode changes to classe 731 * htb_do_events - make mode changes to classes at the level
738 * @q: the priority event queue 732 * @q: the priority event queue
739 * @level: which wait_pq in 'q->hlevel' 733 * @level: which wait_pq in 'q->hlevel'
740 * @start: start jiffies 734 * @start: start jiffies
741 * 735 *
742 * Scans event queue for pending events and ap 736 * Scans event queue for pending events and applies them. Returns time of
743 * next pending event (0 for no event in pq, q 737 * next pending event (0 for no event in pq, q->now for too many events).
744 * Note: Applied are events whose have cl->pq_ 738 * Note: Applied are events whose have cl->pq_key <= q->now.
745 */ 739 */
746 static s64 htb_do_events(struct htb_sched *q, 740 static s64 htb_do_events(struct htb_sched *q, const int level,
747 unsigned long start) 741 unsigned long start)
748 { 742 {
749 /* don't run for longer than 2 jiffies 743 /* don't run for longer than 2 jiffies; 2 is used instead of
750 * 1 to simplify things when jiffy is 744 * 1 to simplify things when jiffy is going to be incremented
751 * too soon 745 * too soon
752 */ 746 */
753 unsigned long stop_at = start + 2; 747 unsigned long stop_at = start + 2;
754 struct rb_root *wait_pq = &q->hlevel[l 748 struct rb_root *wait_pq = &q->hlevel[level].wait_pq;
755 749
756 while (time_before(jiffies, stop_at)) 750 while (time_before(jiffies, stop_at)) {
757 struct htb_class *cl; 751 struct htb_class *cl;
758 s64 diff; 752 s64 diff;
759 struct rb_node *p = rb_first(w 753 struct rb_node *p = rb_first(wait_pq);
760 754
761 if (!p) 755 if (!p)
762 return 0; 756 return 0;
763 757
764 cl = rb_entry(p, struct htb_cl 758 cl = rb_entry(p, struct htb_class, pq_node);
765 if (cl->pq_key > q->now) 759 if (cl->pq_key > q->now)
766 return cl->pq_key; 760 return cl->pq_key;
767 761
768 htb_safe_rb_erase(p, wait_pq); 762 htb_safe_rb_erase(p, wait_pq);
769 diff = min_t(s64, q->now - cl- 763 diff = min_t(s64, q->now - cl->t_c, cl->mbuffer);
770 htb_change_class_mode(q, cl, & 764 htb_change_class_mode(q, cl, &diff);
771 if (cl->cmode != HTB_CAN_SEND) 765 if (cl->cmode != HTB_CAN_SEND)
772 htb_add_to_wait_tree(q 766 htb_add_to_wait_tree(q, cl, diff);
773 } 767 }
774 768
775 /* too much load - let's continue afte 769 /* too much load - let's continue after a break for scheduling */
776 if (!(q->warned & HTB_WARN_TOOMANYEVEN 770 if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) {
777 pr_warn("htb: too many events! 771 pr_warn("htb: too many events!\n");
778 q->warned |= HTB_WARN_TOOMANYE 772 q->warned |= HTB_WARN_TOOMANYEVENTS;
779 } 773 }
780 774
781 return q->now; 775 return q->now;
782 } 776 }
783 777
784 /* Returns class->node+prio from id-tree where 778 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
785 * is no such one exists. 779 * is no such one exists.
786 */ 780 */
787 static struct rb_node *htb_id_find_next_upper( 781 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n,
788 782 u32 id)
789 { 783 {
790 struct rb_node *r = NULL; 784 struct rb_node *r = NULL;
791 while (n) { 785 while (n) {
792 struct htb_class *cl = 786 struct htb_class *cl =
793 rb_entry(n, struct htb_cla 787 rb_entry(n, struct htb_class, node[prio]);
794 788
795 if (id > cl->common.classid) { 789 if (id > cl->common.classid) {
796 n = n->rb_right; 790 n = n->rb_right;
797 } else if (id < cl->common.cla 791 } else if (id < cl->common.classid) {
798 r = n; 792 r = n;
799 n = n->rb_left; 793 n = n->rb_left;
800 } else { 794 } else {
801 return n; 795 return n;
802 } 796 }
803 } 797 }
804 return r; 798 return r;
805 } 799 }
806 800
807 /** 801 /**
808 * htb_lookup_leaf - returns next leaf class i 802 * htb_lookup_leaf - returns next leaf class in DRR order
809 * @hprio: the current one 803 * @hprio: the current one
810 * @prio: which prio in class 804 * @prio: which prio in class
811 * 805 *
812 * Find leaf where current feed pointers point 806 * Find leaf where current feed pointers points to.
813 */ 807 */
814 static struct htb_class *htb_lookup_leaf(struc 808 static struct htb_class *htb_lookup_leaf(struct htb_prio *hprio, const int prio)
815 { 809 {
816 int i; 810 int i;
817 struct { 811 struct {
818 struct rb_node *root; 812 struct rb_node *root;
819 struct rb_node **pptr; 813 struct rb_node **pptr;
820 u32 *pid; 814 u32 *pid;
821 } stk[TC_HTB_MAXDEPTH], *sp = stk; 815 } stk[TC_HTB_MAXDEPTH], *sp = stk;
822 816
823 BUG_ON(!hprio->row.rb_node); 817 BUG_ON(!hprio->row.rb_node);
824 sp->root = hprio->row.rb_node; 818 sp->root = hprio->row.rb_node;
825 sp->pptr = &hprio->ptr; 819 sp->pptr = &hprio->ptr;
826 sp->pid = &hprio->last_ptr_id; 820 sp->pid = &hprio->last_ptr_id;
827 821
828 for (i = 0; i < 65535; i++) { 822 for (i = 0; i < 65535; i++) {
829 if (!*sp->pptr && *sp->pid) { 823 if (!*sp->pptr && *sp->pid) {
830 /* ptr was invalidated 824 /* ptr was invalidated but id is valid - try to recover
831 * the original or nex 825 * the original or next ptr
832 */ 826 */
833 *sp->pptr = 827 *sp->pptr =
834 htb_id_find_next_u 828 htb_id_find_next_upper(prio, sp->root, *sp->pid);
835 } 829 }
836 *sp->pid = 0; /* ptr is vali 830 *sp->pid = 0; /* ptr is valid now so that remove this hint as it
837 * can become 831 * can become out of date quickly
838 */ 832 */
839 if (!*sp->pptr) { /* we 833 if (!*sp->pptr) { /* we are at right end; rewind & go up */
840 *sp->pptr = sp->root; 834 *sp->pptr = sp->root;
841 while ((*sp->pptr)->rb 835 while ((*sp->pptr)->rb_left)
842 *sp->pptr = (* 836 *sp->pptr = (*sp->pptr)->rb_left;
843 if (sp > stk) { 837 if (sp > stk) {
844 sp--; 838 sp--;
845 if (!*sp->pptr 839 if (!*sp->pptr) {
846 WARN_O 840 WARN_ON(1);
847 return 841 return NULL;
848 } 842 }
849 htb_next_rb_no 843 htb_next_rb_node(sp->pptr);
850 } 844 }
851 } else { 845 } else {
852 struct htb_class *cl; 846 struct htb_class *cl;
853 struct htb_prio *clp; 847 struct htb_prio *clp;
854 848
855 cl = rb_entry(*sp->ppt 849 cl = rb_entry(*sp->pptr, struct htb_class, node[prio]);
856 if (!cl->level) 850 if (!cl->level)
857 return cl; 851 return cl;
858 clp = &cl->inner.clpri 852 clp = &cl->inner.clprio[prio];
859 (++sp)->root = clp->fe 853 (++sp)->root = clp->feed.rb_node;
860 sp->pptr = &clp->ptr; 854 sp->pptr = &clp->ptr;
861 sp->pid = &clp->last_p 855 sp->pid = &clp->last_ptr_id;
862 } 856 }
863 } 857 }
864 WARN_ON(1); 858 WARN_ON(1);
865 return NULL; 859 return NULL;
866 } 860 }
867 861
868 /* dequeues packet at given priority and level 862 /* dequeues packet at given priority and level; call only if
869 * you are sure that there is active class at 863 * you are sure that there is active class at prio/level
870 */ 864 */
871 static struct sk_buff *htb_dequeue_tree(struct 865 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, const int prio,
872 const 866 const int level)
873 { 867 {
874 struct sk_buff *skb = NULL; 868 struct sk_buff *skb = NULL;
875 struct htb_class *cl, *start; 869 struct htb_class *cl, *start;
876 struct htb_level *hlevel = &q->hlevel[ 870 struct htb_level *hlevel = &q->hlevel[level];
877 struct htb_prio *hprio = &hlevel->hpri 871 struct htb_prio *hprio = &hlevel->hprio[prio];
878 872
879 /* look initial class up in the row */ 873 /* look initial class up in the row */
880 start = cl = htb_lookup_leaf(hprio, pr 874 start = cl = htb_lookup_leaf(hprio, prio);
881 875
882 do { 876 do {
883 next: 877 next:
884 if (unlikely(!cl)) 878 if (unlikely(!cl))
885 return NULL; 879 return NULL;
886 880
887 /* class can be empty - it is 881 /* class can be empty - it is unlikely but can be true if leaf
888 * qdisc drops packets in enqu 882 * qdisc drops packets in enqueue routine or if someone used
889 * graft operation on the leaf 883 * graft operation on the leaf since last dequeue;
890 * simply deactivate and skip 884 * simply deactivate and skip such class
891 */ 885 */
892 if (unlikely(cl->leaf.q->q.qle 886 if (unlikely(cl->leaf.q->q.qlen == 0)) {
893 struct htb_class *next 887 struct htb_class *next;
894 htb_deactivate(q, cl); 888 htb_deactivate(q, cl);
895 889
896 /* row/level might bec 890 /* row/level might become empty */
897 if ((q->row_mask[level 891 if ((q->row_mask[level] & (1 << prio)) == 0)
898 return NULL; 892 return NULL;
899 893
900 next = htb_lookup_leaf 894 next = htb_lookup_leaf(hprio, prio);
901 895
902 if (cl == start) 896 if (cl == start) /* fix start if we just deleted it */
903 start = next; 897 start = next;
904 cl = next; 898 cl = next;
905 goto next; 899 goto next;
906 } 900 }
907 901
908 skb = cl->leaf.q->dequeue(cl-> 902 skb = cl->leaf.q->dequeue(cl->leaf.q);
909 if (likely(skb != NULL)) 903 if (likely(skb != NULL))
910 break; 904 break;
911 905
912 qdisc_warn_nonwc("htb", cl->le 906 qdisc_warn_nonwc("htb", cl->leaf.q);
913 htb_next_rb_node(level ? &cl-> 907 htb_next_rb_node(level ? &cl->parent->inner.clprio[prio].ptr:
914 &q->h 908 &q->hlevel[0].hprio[prio].ptr);
915 cl = htb_lookup_leaf(hprio, pr 909 cl = htb_lookup_leaf(hprio, prio);
916 910
917 } while (cl != start); 911 } while (cl != start);
918 912
919 if (likely(skb != NULL)) { 913 if (likely(skb != NULL)) {
920 bstats_update(&cl->bstats, skb 914 bstats_update(&cl->bstats, skb);
921 cl->leaf.deficit[level] -= qdi 915 cl->leaf.deficit[level] -= qdisc_pkt_len(skb);
922 if (cl->leaf.deficit[level] < 916 if (cl->leaf.deficit[level] < 0) {
923 cl->leaf.deficit[level 917 cl->leaf.deficit[level] += cl->quantum;
924 htb_next_rb_node(level 918 htb_next_rb_node(level ? &cl->parent->inner.clprio[prio].ptr :
925 919 &q->hlevel[0].hprio[prio].ptr);
926 } 920 }
927 /* this used to be after charg 921 /* this used to be after charge_class but this constelation
928 * gives us slightly better pe 922 * gives us slightly better performance
929 */ 923 */
930 if (!cl->leaf.q->q.qlen) 924 if (!cl->leaf.q->q.qlen)
931 htb_deactivate(q, cl); 925 htb_deactivate(q, cl);
932 htb_charge_class(q, cl, level, 926 htb_charge_class(q, cl, level, skb);
933 } 927 }
934 return skb; 928 return skb;
935 } 929 }
936 930
937 static struct sk_buff *htb_dequeue(struct Qdis 931 static struct sk_buff *htb_dequeue(struct Qdisc *sch)
938 { 932 {
939 struct sk_buff *skb; 933 struct sk_buff *skb;
940 struct htb_sched *q = qdisc_priv(sch); 934 struct htb_sched *q = qdisc_priv(sch);
941 int level; 935 int level;
942 s64 next_event; 936 s64 next_event;
943 unsigned long start_at; 937 unsigned long start_at;
944 938
945 /* try to dequeue direct packets as hi 939 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
946 skb = __qdisc_dequeue_head(&q->direct_ 940 skb = __qdisc_dequeue_head(&q->direct_queue);
947 if (skb != NULL) { 941 if (skb != NULL) {
948 ok: 942 ok:
949 qdisc_bstats_update(sch, skb); 943 qdisc_bstats_update(sch, skb);
950 qdisc_qstats_backlog_dec(sch, 944 qdisc_qstats_backlog_dec(sch, skb);
951 sch->q.qlen--; 945 sch->q.qlen--;
952 return skb; 946 return skb;
953 } 947 }
954 948
955 if (!sch->q.qlen) 949 if (!sch->q.qlen)
956 goto fin; 950 goto fin;
957 q->now = ktime_get_ns(); 951 q->now = ktime_get_ns();
958 start_at = jiffies; 952 start_at = jiffies;
959 953
960 next_event = q->now + 5LLU * NSEC_PER_ 954 next_event = q->now + 5LLU * NSEC_PER_SEC;
961 955
962 for (level = 0; level < TC_HTB_MAXDEPT 956 for (level = 0; level < TC_HTB_MAXDEPTH; level++) {
963 /* common case optimization - 957 /* common case optimization - skip event handler quickly */
964 int m; 958 int m;
965 s64 event = q->near_ev_cache[l 959 s64 event = q->near_ev_cache[level];
966 960
967 if (q->now >= event) { 961 if (q->now >= event) {
968 event = htb_do_events( 962 event = htb_do_events(q, level, start_at);
969 if (!event) 963 if (!event)
970 event = q->now 964 event = q->now + NSEC_PER_SEC;
971 q->near_ev_cache[level 965 q->near_ev_cache[level] = event;
972 } 966 }
973 967
974 if (next_event > event) 968 if (next_event > event)
975 next_event = event; 969 next_event = event;
976 970
977 m = ~q->row_mask[level]; 971 m = ~q->row_mask[level];
978 while (m != (int)(-1)) { 972 while (m != (int)(-1)) {
979 int prio = ffz(m); 973 int prio = ffz(m);
980 974
981 m |= 1 << prio; 975 m |= 1 << prio;
982 skb = htb_dequeue_tree 976 skb = htb_dequeue_tree(q, prio, level);
983 if (likely(skb != NULL 977 if (likely(skb != NULL))
984 goto ok; 978 goto ok;
985 } 979 }
986 } 980 }
987 if (likely(next_event > q->now)) 981 if (likely(next_event > q->now))
988 qdisc_watchdog_schedule_ns(&q- 982 qdisc_watchdog_schedule_ns(&q->watchdog, next_event);
989 else 983 else
990 schedule_work(&q->work); 984 schedule_work(&q->work);
991 fin: 985 fin:
992 return skb; 986 return skb;
993 } 987 }
994 988
995 /* reset all classes */ 989 /* reset all classes */
996 /* always caled under BH & queue lock */ 990 /* always caled under BH & queue lock */
997 static void htb_reset(struct Qdisc *sch) 991 static void htb_reset(struct Qdisc *sch)
998 { 992 {
999 struct htb_sched *q = qdisc_priv(sch); 993 struct htb_sched *q = qdisc_priv(sch);
1000 struct htb_class *cl; 994 struct htb_class *cl;
1001 unsigned int i; 995 unsigned int i;
1002 996
1003 for (i = 0; i < q->clhash.hashsize; i 997 for (i = 0; i < q->clhash.hashsize; i++) {
1004 hlist_for_each_entry(cl, &q-> 998 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
1005 if (cl->level) 999 if (cl->level)
1006 memset(&cl->i 1000 memset(&cl->inner, 0, sizeof(cl->inner));
1007 else { 1001 else {
1008 if (cl->leaf. 1002 if (cl->leaf.q && !q->offload)
1009 qdisc 1003 qdisc_reset(cl->leaf.q);
1010 } 1004 }
1011 cl->prio_activity = 0 1005 cl->prio_activity = 0;
1012 cl->cmode = HTB_CAN_S 1006 cl->cmode = HTB_CAN_SEND;
1013 } 1007 }
1014 } 1008 }
1015 qdisc_watchdog_cancel(&q->watchdog); 1009 qdisc_watchdog_cancel(&q->watchdog);
1016 __qdisc_reset_queue(&q->direct_queue) 1010 __qdisc_reset_queue(&q->direct_queue);
1017 memset(q->hlevel, 0, sizeof(q->hlevel 1011 memset(q->hlevel, 0, sizeof(q->hlevel));
1018 memset(q->row_mask, 0, sizeof(q->row_ 1012 memset(q->row_mask, 0, sizeof(q->row_mask));
1019 } 1013 }
1020 1014
1021 static const struct nla_policy htb_policy[TCA 1015 static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = {
1022 [TCA_HTB_PARMS] = { .len = sizeof(str 1016 [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) },
1023 [TCA_HTB_INIT] = { .len = sizeof(str 1017 [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) },
1024 [TCA_HTB_CTAB] = { .type = NLA_BINAR 1018 [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1025 [TCA_HTB_RTAB] = { .type = NLA_BINAR 1019 [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1026 [TCA_HTB_DIRECT_QLEN] = { .type = NLA 1020 [TCA_HTB_DIRECT_QLEN] = { .type = NLA_U32 },
1027 [TCA_HTB_RATE64] = { .type = NLA_U64 1021 [TCA_HTB_RATE64] = { .type = NLA_U64 },
1028 [TCA_HTB_CEIL64] = { .type = NLA_U64 1022 [TCA_HTB_CEIL64] = { .type = NLA_U64 },
1029 [TCA_HTB_OFFLOAD] = { .type = NLA_FLA 1023 [TCA_HTB_OFFLOAD] = { .type = NLA_FLAG },
1030 }; 1024 };
1031 1025
1032 static void htb_work_func(struct work_struct 1026 static void htb_work_func(struct work_struct *work)
1033 { 1027 {
1034 struct htb_sched *q = container_of(wo 1028 struct htb_sched *q = container_of(work, struct htb_sched, work);
1035 struct Qdisc *sch = q->watchdog.qdisc 1029 struct Qdisc *sch = q->watchdog.qdisc;
1036 1030
1037 rcu_read_lock(); 1031 rcu_read_lock();
1038 __netif_schedule(qdisc_root(sch)); 1032 __netif_schedule(qdisc_root(sch));
1039 rcu_read_unlock(); 1033 rcu_read_unlock();
1040 } 1034 }
1041 1035
>> 1036 static void htb_set_lockdep_class_child(struct Qdisc *q)
>> 1037 {
>> 1038 static struct lock_class_key child_key;
>> 1039
>> 1040 lockdep_set_class(qdisc_lock(q), &child_key);
>> 1041 }
>> 1042
1042 static int htb_offload(struct net_device *dev 1043 static int htb_offload(struct net_device *dev, struct tc_htb_qopt_offload *opt)
1043 { 1044 {
1044 return dev->netdev_ops->ndo_setup_tc( 1045 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_HTB, opt);
1045 } 1046 }
1046 1047
1047 static int htb_init(struct Qdisc *sch, struct 1048 static int htb_init(struct Qdisc *sch, struct nlattr *opt,
1048 struct netlink_ext_ack *e 1049 struct netlink_ext_ack *extack)
1049 { 1050 {
1050 struct net_device *dev = qdisc_dev(sc 1051 struct net_device *dev = qdisc_dev(sch);
1051 struct tc_htb_qopt_offload offload_op 1052 struct tc_htb_qopt_offload offload_opt;
1052 struct htb_sched *q = qdisc_priv(sch) 1053 struct htb_sched *q = qdisc_priv(sch);
1053 struct nlattr *tb[TCA_HTB_MAX + 1]; 1054 struct nlattr *tb[TCA_HTB_MAX + 1];
1054 struct tc_htb_glob *gopt; 1055 struct tc_htb_glob *gopt;
1055 unsigned int ntx; 1056 unsigned int ntx;
1056 bool offload; 1057 bool offload;
1057 int err; 1058 int err;
1058 1059
1059 qdisc_watchdog_init(&q->watchdog, sch 1060 qdisc_watchdog_init(&q->watchdog, sch);
1060 INIT_WORK(&q->work, htb_work_func); 1061 INIT_WORK(&q->work, htb_work_func);
1061 1062
1062 if (!opt) 1063 if (!opt)
1063 return -EINVAL; 1064 return -EINVAL;
1064 1065
1065 err = tcf_block_get(&q->block, &q->fi 1066 err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
1066 if (err) 1067 if (err)
1067 return err; 1068 return err;
1068 1069
1069 err = nla_parse_nested_deprecated(tb, 1070 err = nla_parse_nested_deprecated(tb, TCA_HTB_MAX, opt, htb_policy,
1070 NUL 1071 NULL);
1071 if (err < 0) 1072 if (err < 0)
1072 return err; 1073 return err;
1073 1074
1074 if (!tb[TCA_HTB_INIT]) 1075 if (!tb[TCA_HTB_INIT])
1075 return -EINVAL; 1076 return -EINVAL;
1076 1077
1077 gopt = nla_data(tb[TCA_HTB_INIT]); 1078 gopt = nla_data(tb[TCA_HTB_INIT]);
1078 if (gopt->version != HTB_VER >> 16) 1079 if (gopt->version != HTB_VER >> 16)
1079 return -EINVAL; 1080 return -EINVAL;
1080 1081
1081 offload = nla_get_flag(tb[TCA_HTB_OFF 1082 offload = nla_get_flag(tb[TCA_HTB_OFFLOAD]);
1082 1083
1083 if (offload) { 1084 if (offload) {
1084 if (sch->parent != TC_H_ROOT) 1085 if (sch->parent != TC_H_ROOT) {
1085 NL_SET_ERR_MSG(extack 1086 NL_SET_ERR_MSG(extack, "HTB must be the root qdisc to use offload");
1086 return -EOPNOTSUPP; 1087 return -EOPNOTSUPP;
1087 } 1088 }
1088 1089
1089 if (!tc_can_offload(dev) || ! 1090 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) {
1090 NL_SET_ERR_MSG(extack 1091 NL_SET_ERR_MSG(extack, "hw-tc-offload ethtool feature flag must be on");
1091 return -EOPNOTSUPP; 1092 return -EOPNOTSUPP;
1092 } 1093 }
1093 1094
1094 q->num_direct_qdiscs = dev->r 1095 q->num_direct_qdiscs = dev->real_num_tx_queues;
1095 q->direct_qdiscs = kcalloc(q- 1096 q->direct_qdiscs = kcalloc(q->num_direct_qdiscs,
1096 si 1097 sizeof(*q->direct_qdiscs),
1097 GF 1098 GFP_KERNEL);
1098 if (!q->direct_qdiscs) 1099 if (!q->direct_qdiscs)
1099 return -ENOMEM; 1100 return -ENOMEM;
1100 } 1101 }
1101 1102
1102 err = qdisc_class_hash_init(&q->clhas 1103 err = qdisc_class_hash_init(&q->clhash);
1103 if (err < 0) 1104 if (err < 0)
1104 return err; !! 1105 goto err_free_direct_qdiscs;
>> 1106
>> 1107 qdisc_skb_head_init(&q->direct_queue);
1105 1108
1106 if (tb[TCA_HTB_DIRECT_QLEN]) 1109 if (tb[TCA_HTB_DIRECT_QLEN])
1107 q->direct_qlen = nla_get_u32( 1110 q->direct_qlen = nla_get_u32(tb[TCA_HTB_DIRECT_QLEN]);
1108 else 1111 else
1109 q->direct_qlen = qdisc_dev(sc 1112 q->direct_qlen = qdisc_dev(sch)->tx_queue_len;
1110 1113
1111 if ((q->rate2quantum = gopt->rate2qua 1114 if ((q->rate2quantum = gopt->rate2quantum) < 1)
1112 q->rate2quantum = 1; 1115 q->rate2quantum = 1;
1113 q->defcls = gopt->defcls; 1116 q->defcls = gopt->defcls;
1114 1117
1115 if (!offload) 1118 if (!offload)
1116 return 0; 1119 return 0;
1117 1120
1118 for (ntx = 0; ntx < q->num_direct_qdi 1121 for (ntx = 0; ntx < q->num_direct_qdiscs; ntx++) {
1119 struct netdev_queue *dev_queu 1122 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, ntx);
1120 struct Qdisc *qdisc; 1123 struct Qdisc *qdisc;
1121 1124
1122 qdisc = qdisc_create_dflt(dev 1125 qdisc = qdisc_create_dflt(dev_queue, &pfifo_qdisc_ops,
1123 TC_ 1126 TC_H_MAKE(sch->handle, 0), extack);
1124 if (!qdisc) { 1127 if (!qdisc) {
1125 return -ENOMEM; !! 1128 err = -ENOMEM;
>> 1129 goto err_free_qdiscs;
1126 } 1130 }
1127 1131
>> 1132 htb_set_lockdep_class_child(qdisc);
1128 q->direct_qdiscs[ntx] = qdisc 1133 q->direct_qdiscs[ntx] = qdisc;
1129 qdisc->flags |= TCQ_F_ONETXQU 1134 qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1130 } 1135 }
1131 1136
1132 sch->flags |= TCQ_F_MQROOT; 1137 sch->flags |= TCQ_F_MQROOT;
1133 1138
1134 offload_opt = (struct tc_htb_qopt_off 1139 offload_opt = (struct tc_htb_qopt_offload) {
1135 .command = TC_HTB_CREATE, 1140 .command = TC_HTB_CREATE,
1136 .parent_classid = TC_H_MAJ(sc 1141 .parent_classid = TC_H_MAJ(sch->handle) >> 16,
1137 .classid = TC_H_MIN(q->defcls 1142 .classid = TC_H_MIN(q->defcls),
1138 .extack = extack, 1143 .extack = extack,
1139 }; 1144 };
1140 err = htb_offload(dev, &offload_opt); 1145 err = htb_offload(dev, &offload_opt);
1141 if (err) 1146 if (err)
1142 return err; !! 1147 goto err_free_qdiscs;
1143 1148
1144 /* Defer this assignment, so that htb 1149 /* Defer this assignment, so that htb_destroy skips offload-related
1145 * parts (especially calling ndo_setu 1150 * parts (especially calling ndo_setup_tc) on errors.
1146 */ 1151 */
1147 q->offload = true; 1152 q->offload = true;
1148 1153
1149 return 0; 1154 return 0;
>> 1155
>> 1156 err_free_qdiscs:
>> 1157 for (ntx = 0; ntx < q->num_direct_qdiscs && q->direct_qdiscs[ntx];
>> 1158 ntx++)
>> 1159 qdisc_put(q->direct_qdiscs[ntx]);
>> 1160
>> 1161 qdisc_class_hash_destroy(&q->clhash);
>> 1162 /* Prevent use-after-free and double-free when htb_destroy gets called.
>> 1163 */
>> 1164 q->clhash.hash = NULL;
>> 1165 q->clhash.hashsize = 0;
>> 1166
>> 1167 err_free_direct_qdiscs:
>> 1168 kfree(q->direct_qdiscs);
>> 1169 q->direct_qdiscs = NULL;
>> 1170 return err;
1150 } 1171 }
1151 1172
1152 static void htb_attach_offload(struct Qdisc * 1173 static void htb_attach_offload(struct Qdisc *sch)
1153 { 1174 {
1154 struct net_device *dev = qdisc_dev(sc 1175 struct net_device *dev = qdisc_dev(sch);
1155 struct htb_sched *q = qdisc_priv(sch) 1176 struct htb_sched *q = qdisc_priv(sch);
1156 unsigned int ntx; 1177 unsigned int ntx;
1157 1178
1158 for (ntx = 0; ntx < q->num_direct_qdi 1179 for (ntx = 0; ntx < q->num_direct_qdiscs; ntx++) {
1159 struct Qdisc *old, *qdisc = q 1180 struct Qdisc *old, *qdisc = q->direct_qdiscs[ntx];
1160 1181
1161 old = dev_graft_qdisc(qdisc-> 1182 old = dev_graft_qdisc(qdisc->dev_queue, qdisc);
1162 qdisc_put(old); 1183 qdisc_put(old);
1163 qdisc_hash_add(qdisc, false); 1184 qdisc_hash_add(qdisc, false);
1164 } 1185 }
1165 for (ntx = q->num_direct_qdiscs; ntx 1186 for (ntx = q->num_direct_qdiscs; ntx < dev->num_tx_queues; ntx++) {
1166 struct netdev_queue *dev_queu 1187 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, ntx);
1167 struct Qdisc *old = dev_graft 1188 struct Qdisc *old = dev_graft_qdisc(dev_queue, NULL);
1168 1189
1169 qdisc_put(old); 1190 qdisc_put(old);
1170 } 1191 }
1171 1192
1172 kfree(q->direct_qdiscs); 1193 kfree(q->direct_qdiscs);
1173 q->direct_qdiscs = NULL; 1194 q->direct_qdiscs = NULL;
1174 } 1195 }
1175 1196
1176 static void htb_attach_software(struct Qdisc 1197 static void htb_attach_software(struct Qdisc *sch)
1177 { 1198 {
1178 struct net_device *dev = qdisc_dev(sc 1199 struct net_device *dev = qdisc_dev(sch);
1179 unsigned int ntx; 1200 unsigned int ntx;
1180 1201
1181 /* Resemble qdisc_graft behavior. */ 1202 /* Resemble qdisc_graft behavior. */
1182 for (ntx = 0; ntx < dev->num_tx_queue 1203 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
1183 struct netdev_queue *dev_queu 1204 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, ntx);
1184 struct Qdisc *old = dev_graft 1205 struct Qdisc *old = dev_graft_qdisc(dev_queue, sch);
1185 1206
1186 qdisc_refcount_inc(sch); 1207 qdisc_refcount_inc(sch);
1187 1208
1188 qdisc_put(old); 1209 qdisc_put(old);
1189 } 1210 }
1190 } 1211 }
1191 1212
1192 static void htb_attach(struct Qdisc *sch) 1213 static void htb_attach(struct Qdisc *sch)
1193 { 1214 {
1194 struct htb_sched *q = qdisc_priv(sch) 1215 struct htb_sched *q = qdisc_priv(sch);
1195 1216
1196 if (q->offload) 1217 if (q->offload)
1197 htb_attach_offload(sch); 1218 htb_attach_offload(sch);
1198 else 1219 else
1199 htb_attach_software(sch); 1220 htb_attach_software(sch);
1200 } 1221 }
1201 1222
1202 static int htb_dump(struct Qdisc *sch, struct 1223 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1203 { 1224 {
1204 struct htb_sched *q = qdisc_priv(sch) 1225 struct htb_sched *q = qdisc_priv(sch);
1205 struct nlattr *nest; 1226 struct nlattr *nest;
1206 struct tc_htb_glob gopt; 1227 struct tc_htb_glob gopt;
1207 1228
1208 if (q->offload) 1229 if (q->offload)
1209 sch->flags |= TCQ_F_OFFLOADED 1230 sch->flags |= TCQ_F_OFFLOADED;
1210 else 1231 else
1211 sch->flags &= ~TCQ_F_OFFLOADE 1232 sch->flags &= ~TCQ_F_OFFLOADED;
1212 1233
1213 sch->qstats.overlimits = q->overlimit 1234 sch->qstats.overlimits = q->overlimits;
1214 /* Its safe to not acquire qdisc lock 1235 /* Its safe to not acquire qdisc lock. As we hold RTNL,
1215 * no change can happen on the qdisc 1236 * no change can happen on the qdisc parameters.
1216 */ 1237 */
1217 1238
1218 gopt.direct_pkts = q->direct_pkts; 1239 gopt.direct_pkts = q->direct_pkts;
1219 gopt.version = HTB_VER; 1240 gopt.version = HTB_VER;
1220 gopt.rate2quantum = q->rate2quantum; 1241 gopt.rate2quantum = q->rate2quantum;
1221 gopt.defcls = q->defcls; 1242 gopt.defcls = q->defcls;
1222 gopt.debug = 0; 1243 gopt.debug = 0;
1223 1244
1224 nest = nla_nest_start_noflag(skb, TCA 1245 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1225 if (nest == NULL) 1246 if (nest == NULL)
1226 goto nla_put_failure; 1247 goto nla_put_failure;
1227 if (nla_put(skb, TCA_HTB_INIT, sizeof 1248 if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt) ||
1228 nla_put_u32(skb, TCA_HTB_DIRECT_Q 1249 nla_put_u32(skb, TCA_HTB_DIRECT_QLEN, q->direct_qlen))
1229 goto nla_put_failure; 1250 goto nla_put_failure;
1230 if (q->offload && nla_put_flag(skb, T 1251 if (q->offload && nla_put_flag(skb, TCA_HTB_OFFLOAD))
1231 goto nla_put_failure; 1252 goto nla_put_failure;
1232 1253
1233 return nla_nest_end(skb, nest); 1254 return nla_nest_end(skb, nest);
1234 1255
1235 nla_put_failure: 1256 nla_put_failure:
1236 nla_nest_cancel(skb, nest); 1257 nla_nest_cancel(skb, nest);
1237 return -1; 1258 return -1;
1238 } 1259 }
1239 1260
1240 static int htb_dump_class(struct Qdisc *sch, 1261 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1241 struct sk_buff *skb 1262 struct sk_buff *skb, struct tcmsg *tcm)
1242 { 1263 {
1243 struct htb_class *cl = (struct htb_cl 1264 struct htb_class *cl = (struct htb_class *)arg;
1244 struct htb_sched *q = qdisc_priv(sch) 1265 struct htb_sched *q = qdisc_priv(sch);
1245 struct nlattr *nest; 1266 struct nlattr *nest;
1246 struct tc_htb_opt opt; 1267 struct tc_htb_opt opt;
1247 1268
1248 /* Its safe to not acquire qdisc lock 1269 /* Its safe to not acquire qdisc lock. As we hold RTNL,
1249 * no change can happen on the class 1270 * no change can happen on the class parameters.
1250 */ 1271 */
1251 tcm->tcm_parent = cl->parent ? cl->pa 1272 tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT;
1252 tcm->tcm_handle = cl->common.classid; 1273 tcm->tcm_handle = cl->common.classid;
1253 if (!cl->level && cl->leaf.q) 1274 if (!cl->level && cl->leaf.q)
1254 tcm->tcm_info = cl->leaf.q->h 1275 tcm->tcm_info = cl->leaf.q->handle;
1255 1276
1256 nest = nla_nest_start_noflag(skb, TCA 1277 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1257 if (nest == NULL) 1278 if (nest == NULL)
1258 goto nla_put_failure; 1279 goto nla_put_failure;
1259 1280
1260 memset(&opt, 0, sizeof(opt)); 1281 memset(&opt, 0, sizeof(opt));
1261 1282
1262 psched_ratecfg_getrate(&opt.rate, &cl 1283 psched_ratecfg_getrate(&opt.rate, &cl->rate);
1263 opt.buffer = PSCHED_NS2TICKS(cl->buff 1284 opt.buffer = PSCHED_NS2TICKS(cl->buffer);
1264 psched_ratecfg_getrate(&opt.ceil, &cl 1285 psched_ratecfg_getrate(&opt.ceil, &cl->ceil);
1265 opt.cbuffer = PSCHED_NS2TICKS(cl->cbu 1286 opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer);
1266 opt.quantum = cl->quantum; 1287 opt.quantum = cl->quantum;
1267 opt.prio = cl->prio; 1288 opt.prio = cl->prio;
1268 opt.level = cl->level; 1289 opt.level = cl->level;
1269 if (nla_put(skb, TCA_HTB_PARMS, sizeo 1290 if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt))
1270 goto nla_put_failure; 1291 goto nla_put_failure;
1271 if (q->offload && nla_put_flag(skb, T 1292 if (q->offload && nla_put_flag(skb, TCA_HTB_OFFLOAD))
1272 goto nla_put_failure; 1293 goto nla_put_failure;
1273 if ((cl->rate.rate_bytes_ps >= (1ULL 1294 if ((cl->rate.rate_bytes_ps >= (1ULL << 32)) &&
1274 nla_put_u64_64bit(skb, TCA_HTB_RA 1295 nla_put_u64_64bit(skb, TCA_HTB_RATE64, cl->rate.rate_bytes_ps,
1275 TCA_HTB_PAD)) 1296 TCA_HTB_PAD))
1276 goto nla_put_failure; 1297 goto nla_put_failure;
1277 if ((cl->ceil.rate_bytes_ps >= (1ULL 1298 if ((cl->ceil.rate_bytes_ps >= (1ULL << 32)) &&
1278 nla_put_u64_64bit(skb, TCA_HTB_CE 1299 nla_put_u64_64bit(skb, TCA_HTB_CEIL64, cl->ceil.rate_bytes_ps,
1279 TCA_HTB_PAD)) 1300 TCA_HTB_PAD))
1280 goto nla_put_failure; 1301 goto nla_put_failure;
1281 1302
1282 return nla_nest_end(skb, nest); 1303 return nla_nest_end(skb, nest);
1283 1304
1284 nla_put_failure: 1305 nla_put_failure:
1285 nla_nest_cancel(skb, nest); 1306 nla_nest_cancel(skb, nest);
1286 return -1; 1307 return -1;
1287 } 1308 }
1288 1309
1289 static void htb_offload_aggregate_stats(struc 1310 static void htb_offload_aggregate_stats(struct htb_sched *q,
1290 struc 1311 struct htb_class *cl)
1291 { 1312 {
1292 u64 bytes = 0, packets = 0; 1313 u64 bytes = 0, packets = 0;
1293 struct htb_class *c; 1314 struct htb_class *c;
1294 unsigned int i; 1315 unsigned int i;
1295 1316
1296 gnet_stats_basic_sync_init(&cl->bstat 1317 gnet_stats_basic_sync_init(&cl->bstats);
1297 1318
1298 for (i = 0; i < q->clhash.hashsize; i 1319 for (i = 0; i < q->clhash.hashsize; i++) {
1299 hlist_for_each_entry(c, &q->c 1320 hlist_for_each_entry(c, &q->clhash.hash[i], common.hnode) {
1300 struct htb_class *p = 1321 struct htb_class *p = c;
1301 1322
1302 while (p && p->level 1323 while (p && p->level < cl->level)
1303 p = p->parent 1324 p = p->parent;
1304 1325
1305 if (p != cl) 1326 if (p != cl)
1306 continue; 1327 continue;
1307 1328
1308 bytes += u64_stats_re 1329 bytes += u64_stats_read(&c->bstats_bias.bytes);
1309 packets += u64_stats_ 1330 packets += u64_stats_read(&c->bstats_bias.packets);
1310 if (c->level == 0) { 1331 if (c->level == 0) {
1311 bytes += u64_ 1332 bytes += u64_stats_read(&c->leaf.q->bstats.bytes);
1312 packets += u6 1333 packets += u64_stats_read(&c->leaf.q->bstats.packets);
1313 } 1334 }
1314 } 1335 }
1315 } 1336 }
1316 _bstats_update(&cl->bstats, bytes, pa 1337 _bstats_update(&cl->bstats, bytes, packets);
1317 } 1338 }
1318 1339
1319 static int 1340 static int
1320 htb_dump_class_stats(struct Qdisc *sch, unsig 1341 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d)
1321 { 1342 {
1322 struct htb_class *cl = (struct htb_cl 1343 struct htb_class *cl = (struct htb_class *)arg;
1323 struct htb_sched *q = qdisc_priv(sch) 1344 struct htb_sched *q = qdisc_priv(sch);
1324 struct gnet_stats_queue qs = { 1345 struct gnet_stats_queue qs = {
1325 .drops = cl->drops, 1346 .drops = cl->drops,
1326 .overlimits = cl->overlimits, 1347 .overlimits = cl->overlimits,
1327 }; 1348 };
1328 __u32 qlen = 0; 1349 __u32 qlen = 0;
1329 1350
1330 if (!cl->level && cl->leaf.q) 1351 if (!cl->level && cl->leaf.q)
1331 qdisc_qstats_qlen_backlog(cl- 1352 qdisc_qstats_qlen_backlog(cl->leaf.q, &qlen, &qs.backlog);
1332 1353
1333 cl->xstats.tokens = clamp_t(s64, PSCH 1354 cl->xstats.tokens = clamp_t(s64, PSCHED_NS2TICKS(cl->tokens),
1334 INT_MIN, 1355 INT_MIN, INT_MAX);
1335 cl->xstats.ctokens = clamp_t(s64, PSC 1356 cl->xstats.ctokens = clamp_t(s64, PSCHED_NS2TICKS(cl->ctokens),
1336 INT_MIN, 1357 INT_MIN, INT_MAX);
1337 1358
1338 if (q->offload) { 1359 if (q->offload) {
1339 if (!cl->level) { 1360 if (!cl->level) {
1340 if (cl->leaf.q) 1361 if (cl->leaf.q)
1341 cl->bstats = 1362 cl->bstats = cl->leaf.q->bstats;
1342 else 1363 else
1343 gnet_stats_ba 1364 gnet_stats_basic_sync_init(&cl->bstats);
1344 _bstats_update(&cl->b 1365 _bstats_update(&cl->bstats,
1345 u64_st 1366 u64_stats_read(&cl->bstats_bias.bytes),
1346 u64_st 1367 u64_stats_read(&cl->bstats_bias.packets));
1347 } else { 1368 } else {
1348 htb_offload_aggregate 1369 htb_offload_aggregate_stats(q, cl);
1349 } 1370 }
1350 } 1371 }
1351 1372
1352 if (gnet_stats_copy_basic(d, NULL, &c 1373 if (gnet_stats_copy_basic(d, NULL, &cl->bstats, true) < 0 ||
1353 gnet_stats_copy_rate_est(d, &cl-> 1374 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1354 gnet_stats_copy_queue(d, NULL, &q 1375 gnet_stats_copy_queue(d, NULL, &qs, qlen) < 0)
1355 return -1; 1376 return -1;
1356 1377
1357 return gnet_stats_copy_app(d, &cl->xs 1378 return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1358 } 1379 }
1359 1380
1360 static struct netdev_queue * 1381 static struct netdev_queue *
1361 htb_select_queue(struct Qdisc *sch, struct tc 1382 htb_select_queue(struct Qdisc *sch, struct tcmsg *tcm)
1362 { 1383 {
1363 struct net_device *dev = qdisc_dev(sc 1384 struct net_device *dev = qdisc_dev(sch);
1364 struct tc_htb_qopt_offload offload_op 1385 struct tc_htb_qopt_offload offload_opt;
1365 struct htb_sched *q = qdisc_priv(sch) 1386 struct htb_sched *q = qdisc_priv(sch);
1366 int err; 1387 int err;
1367 1388
1368 if (!q->offload) 1389 if (!q->offload)
1369 return sch->dev_queue; 1390 return sch->dev_queue;
1370 1391
1371 offload_opt = (struct tc_htb_qopt_off 1392 offload_opt = (struct tc_htb_qopt_offload) {
1372 .command = TC_HTB_LEAF_QUERY_ 1393 .command = TC_HTB_LEAF_QUERY_QUEUE,
1373 .classid = TC_H_MIN(tcm->tcm_ 1394 .classid = TC_H_MIN(tcm->tcm_parent),
1374 }; 1395 };
1375 err = htb_offload(dev, &offload_opt); 1396 err = htb_offload(dev, &offload_opt);
1376 if (err || offload_opt.qid >= dev->nu 1397 if (err || offload_opt.qid >= dev->num_tx_queues)
1377 return NULL; 1398 return NULL;
1378 return netdev_get_tx_queue(dev, offlo 1399 return netdev_get_tx_queue(dev, offload_opt.qid);
1379 } 1400 }
1380 1401
1381 static struct Qdisc * 1402 static struct Qdisc *
1382 htb_graft_helper(struct netdev_queue *dev_que 1403 htb_graft_helper(struct netdev_queue *dev_queue, struct Qdisc *new_q)
1383 { 1404 {
1384 struct net_device *dev = dev_queue->d 1405 struct net_device *dev = dev_queue->dev;
1385 struct Qdisc *old_q; 1406 struct Qdisc *old_q;
1386 1407
1387 if (dev->flags & IFF_UP) 1408 if (dev->flags & IFF_UP)
1388 dev_deactivate(dev); 1409 dev_deactivate(dev);
1389 old_q = dev_graft_qdisc(dev_queue, ne 1410 old_q = dev_graft_qdisc(dev_queue, new_q);
1390 if (new_q) 1411 if (new_q)
1391 new_q->flags |= TCQ_F_ONETXQU 1412 new_q->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1392 if (dev->flags & IFF_UP) 1413 if (dev->flags & IFF_UP)
1393 dev_activate(dev); 1414 dev_activate(dev);
1394 1415
1395 return old_q; 1416 return old_q;
1396 } 1417 }
1397 1418
1398 static struct netdev_queue *htb_offload_get_q 1419 static struct netdev_queue *htb_offload_get_queue(struct htb_class *cl)
1399 { 1420 {
1400 struct netdev_queue *queue; 1421 struct netdev_queue *queue;
1401 1422
1402 queue = cl->leaf.offload_queue; 1423 queue = cl->leaf.offload_queue;
1403 if (!(cl->leaf.q->flags & TCQ_F_BUILT 1424 if (!(cl->leaf.q->flags & TCQ_F_BUILTIN))
1404 WARN_ON(cl->leaf.q->dev_queue 1425 WARN_ON(cl->leaf.q->dev_queue != queue);
1405 1426
1406 return queue; 1427 return queue;
1407 } 1428 }
1408 1429
1409 static void htb_offload_move_qdisc(struct Qdi 1430 static void htb_offload_move_qdisc(struct Qdisc *sch, struct htb_class *cl_old,
1410 struct htb 1431 struct htb_class *cl_new, bool destroying)
1411 { 1432 {
1412 struct netdev_queue *queue_old, *queu 1433 struct netdev_queue *queue_old, *queue_new;
1413 struct net_device *dev = qdisc_dev(sc 1434 struct net_device *dev = qdisc_dev(sch);
1414 1435
1415 queue_old = htb_offload_get_queue(cl_ 1436 queue_old = htb_offload_get_queue(cl_old);
1416 queue_new = htb_offload_get_queue(cl_ 1437 queue_new = htb_offload_get_queue(cl_new);
1417 1438
1418 if (!destroying) { 1439 if (!destroying) {
1419 struct Qdisc *qdisc; 1440 struct Qdisc *qdisc;
1420 1441
1421 if (dev->flags & IFF_UP) 1442 if (dev->flags & IFF_UP)
1422 dev_deactivate(dev); 1443 dev_deactivate(dev);
1423 qdisc = dev_graft_qdisc(queue 1444 qdisc = dev_graft_qdisc(queue_old, NULL);
1424 WARN_ON(qdisc != cl_old->leaf 1445 WARN_ON(qdisc != cl_old->leaf.q);
1425 } 1446 }
1426 1447
1427 if (!(cl_old->leaf.q->flags & TCQ_F_B 1448 if (!(cl_old->leaf.q->flags & TCQ_F_BUILTIN))
1428 cl_old->leaf.q->dev_queue = q 1449 cl_old->leaf.q->dev_queue = queue_new;
1429 cl_old->leaf.offload_queue = queue_ne 1450 cl_old->leaf.offload_queue = queue_new;
1430 1451
1431 if (!destroying) { 1452 if (!destroying) {
1432 struct Qdisc *qdisc; 1453 struct Qdisc *qdisc;
1433 1454
1434 qdisc = dev_graft_qdisc(queue 1455 qdisc = dev_graft_qdisc(queue_new, cl_old->leaf.q);
1435 if (dev->flags & IFF_UP) 1456 if (dev->flags & IFF_UP)
1436 dev_activate(dev); 1457 dev_activate(dev);
1437 WARN_ON(!(qdisc->flags & TCQ_ 1458 WARN_ON(!(qdisc->flags & TCQ_F_BUILTIN));
1438 } 1459 }
1439 } 1460 }
1440 1461
1441 static int htb_graft(struct Qdisc *sch, unsig 1462 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1442 struct Qdisc **old, stru 1463 struct Qdisc **old, struct netlink_ext_ack *extack)
1443 { 1464 {
1444 struct netdev_queue *dev_queue = sch- 1465 struct netdev_queue *dev_queue = sch->dev_queue;
1445 struct htb_class *cl = (struct htb_cl 1466 struct htb_class *cl = (struct htb_class *)arg;
1446 struct htb_sched *q = qdisc_priv(sch) 1467 struct htb_sched *q = qdisc_priv(sch);
1447 struct Qdisc *old_q; 1468 struct Qdisc *old_q;
1448 1469
1449 if (cl->level) 1470 if (cl->level)
1450 return -EINVAL; 1471 return -EINVAL;
1451 1472
1452 if (q->offload) 1473 if (q->offload)
1453 dev_queue = htb_offload_get_q 1474 dev_queue = htb_offload_get_queue(cl);
1454 1475
1455 if (!new) { 1476 if (!new) {
1456 new = qdisc_create_dflt(dev_q 1477 new = qdisc_create_dflt(dev_queue, &pfifo_qdisc_ops,
1457 cl->c 1478 cl->common.classid, extack);
1458 if (!new) 1479 if (!new)
1459 return -ENOBUFS; 1480 return -ENOBUFS;
1460 } 1481 }
1461 1482
1462 if (q->offload) { 1483 if (q->offload) {
>> 1484 htb_set_lockdep_class_child(new);
1463 /* One ref for cl->leaf.q, th 1485 /* One ref for cl->leaf.q, the other for dev_queue->qdisc. */
1464 qdisc_refcount_inc(new); 1486 qdisc_refcount_inc(new);
1465 old_q = htb_graft_helper(dev_ 1487 old_q = htb_graft_helper(dev_queue, new);
1466 } 1488 }
1467 1489
1468 *old = qdisc_replace(sch, new, &cl->l 1490 *old = qdisc_replace(sch, new, &cl->leaf.q);
1469 1491
1470 if (q->offload) { 1492 if (q->offload) {
1471 WARN_ON(old_q != *old); 1493 WARN_ON(old_q != *old);
1472 qdisc_put(old_q); 1494 qdisc_put(old_q);
1473 } 1495 }
1474 1496
1475 return 0; 1497 return 0;
1476 } 1498 }
1477 1499
1478 static struct Qdisc *htb_leaf(struct Qdisc *s 1500 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg)
1479 { 1501 {
1480 struct htb_class *cl = (struct htb_cl 1502 struct htb_class *cl = (struct htb_class *)arg;
1481 return !cl->level ? cl->leaf.q : NULL 1503 return !cl->level ? cl->leaf.q : NULL;
1482 } 1504 }
1483 1505
1484 static void htb_qlen_notify(struct Qdisc *sch 1506 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg)
1485 { 1507 {
1486 struct htb_class *cl = (struct htb_cl 1508 struct htb_class *cl = (struct htb_class *)arg;
1487 1509
1488 htb_deactivate(qdisc_priv(sch), cl); 1510 htb_deactivate(qdisc_priv(sch), cl);
1489 } 1511 }
1490 1512
1491 static inline int htb_parent_last_child(struc 1513 static inline int htb_parent_last_child(struct htb_class *cl)
1492 { 1514 {
1493 if (!cl->parent) 1515 if (!cl->parent)
1494 /* the root class */ 1516 /* the root class */
1495 return 0; 1517 return 0;
1496 if (cl->parent->children > 1) 1518 if (cl->parent->children > 1)
1497 /* not the last child */ 1519 /* not the last child */
1498 return 0; 1520 return 0;
1499 return 1; 1521 return 1;
1500 } 1522 }
1501 1523
1502 static void htb_parent_to_leaf(struct Qdisc * 1524 static void htb_parent_to_leaf(struct Qdisc *sch, struct htb_class *cl,
1503 struct Qdisc * 1525 struct Qdisc *new_q)
1504 { 1526 {
1505 struct htb_sched *q = qdisc_priv(sch) 1527 struct htb_sched *q = qdisc_priv(sch);
1506 struct htb_class *parent = cl->parent 1528 struct htb_class *parent = cl->parent;
1507 1529
1508 WARN_ON(cl->level || !cl->leaf.q || c 1530 WARN_ON(cl->level || !cl->leaf.q || cl->prio_activity);
1509 1531
1510 if (parent->cmode != HTB_CAN_SEND) 1532 if (parent->cmode != HTB_CAN_SEND)
1511 htb_safe_rb_erase(&parent->pq 1533 htb_safe_rb_erase(&parent->pq_node,
1512 &q->hlevel[ 1534 &q->hlevel[parent->level].wait_pq);
1513 1535
1514 parent->level = 0; 1536 parent->level = 0;
1515 memset(&parent->inner, 0, sizeof(pare 1537 memset(&parent->inner, 0, sizeof(parent->inner));
1516 parent->leaf.q = new_q ? new_q : &noo 1538 parent->leaf.q = new_q ? new_q : &noop_qdisc;
1517 parent->tokens = parent->buffer; 1539 parent->tokens = parent->buffer;
1518 parent->ctokens = parent->cbuffer; 1540 parent->ctokens = parent->cbuffer;
1519 parent->t_c = ktime_get_ns(); 1541 parent->t_c = ktime_get_ns();
1520 parent->cmode = HTB_CAN_SEND; 1542 parent->cmode = HTB_CAN_SEND;
1521 if (q->offload) 1543 if (q->offload)
1522 parent->leaf.offload_queue = 1544 parent->leaf.offload_queue = cl->leaf.offload_queue;
1523 } 1545 }
1524 1546
1525 static void htb_parent_to_leaf_offload(struct 1547 static void htb_parent_to_leaf_offload(struct Qdisc *sch,
1526 struct 1548 struct netdev_queue *dev_queue,
1527 struct 1549 struct Qdisc *new_q)
1528 { 1550 {
1529 struct Qdisc *old_q; 1551 struct Qdisc *old_q;
1530 1552
1531 /* One ref for cl->leaf.q, the other 1553 /* One ref for cl->leaf.q, the other for dev_queue->qdisc. */
1532 if (new_q) 1554 if (new_q)
1533 qdisc_refcount_inc(new_q); 1555 qdisc_refcount_inc(new_q);
1534 old_q = htb_graft_helper(dev_queue, n 1556 old_q = htb_graft_helper(dev_queue, new_q);
1535 WARN_ON(!(old_q->flags & TCQ_F_BUILTI 1557 WARN_ON(!(old_q->flags & TCQ_F_BUILTIN));
1536 } 1558 }
1537 1559
1538 static int htb_destroy_class_offload(struct Q 1560 static int htb_destroy_class_offload(struct Qdisc *sch, struct htb_class *cl,
1539 bool las 1561 bool last_child, bool destroying,
1540 struct n 1562 struct netlink_ext_ack *extack)
1541 { 1563 {
1542 struct tc_htb_qopt_offload offload_op 1564 struct tc_htb_qopt_offload offload_opt;
1543 struct netdev_queue *dev_queue; 1565 struct netdev_queue *dev_queue;
1544 struct Qdisc *q = cl->leaf.q; 1566 struct Qdisc *q = cl->leaf.q;
1545 struct Qdisc *old; !! 1567 struct Qdisc *old = NULL;
1546 int err; 1568 int err;
1547 1569
1548 if (cl->level) 1570 if (cl->level)
1549 return -EINVAL; 1571 return -EINVAL;
1550 1572
1551 WARN_ON(!q); 1573 WARN_ON(!q);
1552 dev_queue = htb_offload_get_queue(cl) 1574 dev_queue = htb_offload_get_queue(cl);
1553 /* When destroying, caller qdisc_graf !! 1575 old = htb_graft_helper(dev_queue, NULL);
1554 * qdisc_put for the qdisc being dest !! 1576 if (destroying)
1555 * does not need to graft or qdisc_pu !! 1577 /* Before HTB is destroyed, the kernel grafts noop_qdisc to
1556 */ !! 1578 * all queues.
1557 if (!destroying) { <<
1558 old = htb_graft_helper(dev_qu <<
1559 /* Last qdisc grafted should <<
1560 * calling htb_delete. <<
1561 */ 1579 */
>> 1580 WARN_ON(!(old->flags & TCQ_F_BUILTIN));
>> 1581 else
1562 WARN_ON(old != q); 1582 WARN_ON(old != q);
1563 } <<
1564 1583
1565 if (cl->parent) { 1584 if (cl->parent) {
1566 _bstats_update(&cl->parent->b 1585 _bstats_update(&cl->parent->bstats_bias,
1567 u64_stats_read 1586 u64_stats_read(&q->bstats.bytes),
1568 u64_stats_read 1587 u64_stats_read(&q->bstats.packets));
1569 } 1588 }
1570 1589
1571 offload_opt = (struct tc_htb_qopt_off 1590 offload_opt = (struct tc_htb_qopt_offload) {
1572 .command = !last_child ? TC_H 1591 .command = !last_child ? TC_HTB_LEAF_DEL :
1573 destroying ? TC_HT 1592 destroying ? TC_HTB_LEAF_DEL_LAST_FORCE :
1574 TC_HTB_LEAF_DEL_LA 1593 TC_HTB_LEAF_DEL_LAST,
1575 .classid = cl->common.classid 1594 .classid = cl->common.classid,
1576 .extack = extack, 1595 .extack = extack,
1577 }; 1596 };
1578 err = htb_offload(qdisc_dev(sch), &of 1597 err = htb_offload(qdisc_dev(sch), &offload_opt);
1579 1598
1580 if (!destroying) { !! 1599 if (!err || destroying)
1581 if (!err) !! 1600 qdisc_put(old);
1582 qdisc_put(old); !! 1601 else
1583 else !! 1602 htb_graft_helper(dev_queue, old);
1584 htb_graft_helper(dev_ <<
1585 } <<
1586 1603
1587 if (last_child) 1604 if (last_child)
1588 return err; 1605 return err;
1589 1606
1590 if (!err && offload_opt.classid != TC 1607 if (!err && offload_opt.classid != TC_H_MIN(cl->common.classid)) {
1591 u32 classid = TC_H_MAJ(sch->h 1608 u32 classid = TC_H_MAJ(sch->handle) |
1592 TC_H_MIN(offloa 1609 TC_H_MIN(offload_opt.classid);
1593 struct htb_class *moved_cl = 1610 struct htb_class *moved_cl = htb_find(classid, sch);
1594 1611
1595 htb_offload_move_qdisc(sch, m 1612 htb_offload_move_qdisc(sch, moved_cl, cl, destroying);
1596 } 1613 }
1597 1614
1598 return err; 1615 return err;
1599 } 1616 }
1600 1617
1601 static void htb_destroy_class(struct Qdisc *s 1618 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
1602 { 1619 {
1603 if (!cl->level) { 1620 if (!cl->level) {
1604 WARN_ON(!cl->leaf.q); 1621 WARN_ON(!cl->leaf.q);
1605 qdisc_put(cl->leaf.q); 1622 qdisc_put(cl->leaf.q);
1606 } 1623 }
1607 gen_kill_estimator(&cl->rate_est); 1624 gen_kill_estimator(&cl->rate_est);
1608 tcf_block_put(cl->block); 1625 tcf_block_put(cl->block);
1609 kfree(cl); 1626 kfree(cl);
1610 } 1627 }
1611 1628
1612 static void htb_destroy(struct Qdisc *sch) 1629 static void htb_destroy(struct Qdisc *sch)
1613 { 1630 {
1614 struct net_device *dev = qdisc_dev(sc 1631 struct net_device *dev = qdisc_dev(sch);
1615 struct tc_htb_qopt_offload offload_op 1632 struct tc_htb_qopt_offload offload_opt;
1616 struct htb_sched *q = qdisc_priv(sch) 1633 struct htb_sched *q = qdisc_priv(sch);
1617 struct hlist_node *next; 1634 struct hlist_node *next;
1618 bool nonempty, changed; 1635 bool nonempty, changed;
1619 struct htb_class *cl; 1636 struct htb_class *cl;
1620 unsigned int i; 1637 unsigned int i;
1621 1638
1622 cancel_work_sync(&q->work); 1639 cancel_work_sync(&q->work);
1623 qdisc_watchdog_cancel(&q->watchdog); 1640 qdisc_watchdog_cancel(&q->watchdog);
1624 /* This line used to be after htb_des 1641 /* This line used to be after htb_destroy_class call below
1625 * and surprisingly it worked in 2.4. 1642 * and surprisingly it worked in 2.4. But it must precede it
1626 * because filter need its target cla 1643 * because filter need its target class alive to be able to call
1627 * unbind_filter on it (without Oops) 1644 * unbind_filter on it (without Oops).
1628 */ 1645 */
1629 tcf_block_put(q->block); 1646 tcf_block_put(q->block);
1630 1647
1631 for (i = 0; i < q->clhash.hashsize; i 1648 for (i = 0; i < q->clhash.hashsize; i++) {
1632 hlist_for_each_entry(cl, &q-> 1649 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
1633 tcf_block_put(cl->blo 1650 tcf_block_put(cl->block);
1634 cl->block = NULL; 1651 cl->block = NULL;
1635 } 1652 }
1636 } 1653 }
1637 1654
1638 do { 1655 do {
1639 nonempty = false; 1656 nonempty = false;
1640 changed = false; 1657 changed = false;
1641 for (i = 0; i < q->clhash.has 1658 for (i = 0; i < q->clhash.hashsize; i++) {
1642 hlist_for_each_entry_ 1659 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
1643 1660 common.hnode) {
1644 bool last_chi 1661 bool last_child;
1645 1662
1646 if (!q->offlo 1663 if (!q->offload) {
1647 htb_d 1664 htb_destroy_class(sch, cl);
1648 conti 1665 continue;
1649 } 1666 }
1650 1667
1651 nonempty = tr 1668 nonempty = true;
1652 1669
1653 if (cl->level 1670 if (cl->level)
1654 conti 1671 continue;
1655 1672
1656 changed = tru 1673 changed = true;
1657 1674
1658 last_child = 1675 last_child = htb_parent_last_child(cl);
1659 htb_destroy_c 1676 htb_destroy_class_offload(sch, cl, last_child,
1660 1677 true, NULL);
1661 qdisc_class_h 1678 qdisc_class_hash_remove(&q->clhash,
1662 1679 &cl->common);
1663 if (cl->paren 1680 if (cl->parent)
1664 cl->p 1681 cl->parent->children--;
1665 if (last_chil 1682 if (last_child)
1666 htb_p 1683 htb_parent_to_leaf(sch, cl, NULL);
1667 htb_destroy_c 1684 htb_destroy_class(sch, cl);
1668 } 1685 }
1669 } 1686 }
1670 } while (changed); 1687 } while (changed);
1671 WARN_ON(nonempty); 1688 WARN_ON(nonempty);
1672 1689
1673 qdisc_class_hash_destroy(&q->clhash); 1690 qdisc_class_hash_destroy(&q->clhash);
1674 __qdisc_reset_queue(&q->direct_queue) 1691 __qdisc_reset_queue(&q->direct_queue);
1675 1692
1676 if (q->offload) { !! 1693 if (!q->offload)
1677 offload_opt = (struct tc_htb_ !! 1694 return;
1678 .command = TC_HTB_DES !! 1695
1679 }; !! 1696 offload_opt = (struct tc_htb_qopt_offload) {
1680 htb_offload(dev, &offload_opt !! 1697 .command = TC_HTB_DESTROY,
1681 } !! 1698 };
>> 1699 htb_offload(dev, &offload_opt);
1682 1700
1683 if (!q->direct_qdiscs) 1701 if (!q->direct_qdiscs)
1684 return; 1702 return;
1685 for (i = 0; i < q->num_direct_qdiscs 1703 for (i = 0; i < q->num_direct_qdiscs && q->direct_qdiscs[i]; i++)
1686 qdisc_put(q->direct_qdiscs[i] 1704 qdisc_put(q->direct_qdiscs[i]);
1687 kfree(q->direct_qdiscs); 1705 kfree(q->direct_qdiscs);
1688 } 1706 }
1689 1707
1690 static int htb_delete(struct Qdisc *sch, unsi 1708 static int htb_delete(struct Qdisc *sch, unsigned long arg,
1691 struct netlink_ext_ack 1709 struct netlink_ext_ack *extack)
1692 { 1710 {
1693 struct htb_sched *q = qdisc_priv(sch) 1711 struct htb_sched *q = qdisc_priv(sch);
1694 struct htb_class *cl = (struct htb_cl 1712 struct htb_class *cl = (struct htb_class *)arg;
1695 struct Qdisc *new_q = NULL; 1713 struct Qdisc *new_q = NULL;
1696 int last_child = 0; 1714 int last_child = 0;
1697 int err; 1715 int err;
1698 1716
1699 /* TODO: why don't allow to delete su 1717 /* TODO: why don't allow to delete subtree ? references ? does
1700 * tc subsys guarantee us that in htb 1718 * tc subsys guarantee us that in htb_destroy it holds no class
1701 * refs so that we can remove childre 1719 * refs so that we can remove children safely there ?
1702 */ 1720 */
1703 if (cl->children || qdisc_class_in_us !! 1721 if (cl->children || cl->filter_cnt)
1704 NL_SET_ERR_MSG(extack, "HTB c <<
1705 return -EBUSY; 1722 return -EBUSY;
1706 } <<
1707 1723
1708 if (!cl->level && htb_parent_last_chi 1724 if (!cl->level && htb_parent_last_child(cl))
1709 last_child = 1; 1725 last_child = 1;
1710 1726
1711 if (q->offload) { 1727 if (q->offload) {
1712 err = htb_destroy_class_offlo 1728 err = htb_destroy_class_offload(sch, cl, last_child, false,
1713 1729 extack);
1714 if (err) 1730 if (err)
1715 return err; 1731 return err;
1716 } 1732 }
1717 1733
1718 if (last_child) { 1734 if (last_child) {
1719 struct netdev_queue *dev_queu 1735 struct netdev_queue *dev_queue = sch->dev_queue;
1720 1736
1721 if (q->offload) 1737 if (q->offload)
1722 dev_queue = htb_offlo 1738 dev_queue = htb_offload_get_queue(cl);
1723 1739
1724 new_q = qdisc_create_dflt(dev 1740 new_q = qdisc_create_dflt(dev_queue, &pfifo_qdisc_ops,
1725 cl- 1741 cl->parent->common.classid,
1726 NUL 1742 NULL);
1727 if (q->offload) !! 1743 if (q->offload) {
>> 1744 if (new_q)
>> 1745 htb_set_lockdep_class_child(new_q);
1728 htb_parent_to_leaf_of 1746 htb_parent_to_leaf_offload(sch, dev_queue, new_q);
>> 1747 }
1729 } 1748 }
1730 1749
1731 sch_tree_lock(sch); 1750 sch_tree_lock(sch);
1732 1751
1733 if (!cl->level) 1752 if (!cl->level)
1734 qdisc_purge_queue(cl->leaf.q) 1753 qdisc_purge_queue(cl->leaf.q);
1735 1754
1736 /* delete from hash and active; remai 1755 /* delete from hash and active; remainder in destroy_class */
1737 qdisc_class_hash_remove(&q->clhash, & 1756 qdisc_class_hash_remove(&q->clhash, &cl->common);
1738 if (cl->parent) 1757 if (cl->parent)
1739 cl->parent->children--; 1758 cl->parent->children--;
1740 1759
1741 if (cl->prio_activity) 1760 if (cl->prio_activity)
1742 htb_deactivate(q, cl); 1761 htb_deactivate(q, cl);
1743 1762
1744 if (cl->cmode != HTB_CAN_SEND) 1763 if (cl->cmode != HTB_CAN_SEND)
1745 htb_safe_rb_erase(&cl->pq_nod 1764 htb_safe_rb_erase(&cl->pq_node,
1746 &q->hlevel[ 1765 &q->hlevel[cl->level].wait_pq);
1747 1766
1748 if (last_child) 1767 if (last_child)
1749 htb_parent_to_leaf(sch, cl, n 1768 htb_parent_to_leaf(sch, cl, new_q);
1750 1769
1751 sch_tree_unlock(sch); 1770 sch_tree_unlock(sch);
1752 1771
1753 htb_destroy_class(sch, cl); 1772 htb_destroy_class(sch, cl);
1754 return 0; 1773 return 0;
1755 } 1774 }
1756 1775
1757 static int htb_change_class(struct Qdisc *sch 1776 static int htb_change_class(struct Qdisc *sch, u32 classid,
1758 u32 parentid, str 1777 u32 parentid, struct nlattr **tca,
1759 unsigned long *ar 1778 unsigned long *arg, struct netlink_ext_ack *extack)
1760 { 1779 {
1761 int err = -EINVAL; 1780 int err = -EINVAL;
1762 struct htb_sched *q = qdisc_priv(sch) 1781 struct htb_sched *q = qdisc_priv(sch);
1763 struct htb_class *cl = (struct htb_cl 1782 struct htb_class *cl = (struct htb_class *)*arg, *parent;
1764 struct tc_htb_qopt_offload offload_op 1783 struct tc_htb_qopt_offload offload_opt;
1765 struct nlattr *opt = tca[TCA_OPTIONS] 1784 struct nlattr *opt = tca[TCA_OPTIONS];
1766 struct nlattr *tb[TCA_HTB_MAX + 1]; 1785 struct nlattr *tb[TCA_HTB_MAX + 1];
1767 struct Qdisc *parent_qdisc = NULL; 1786 struct Qdisc *parent_qdisc = NULL;
1768 struct netdev_queue *dev_queue; 1787 struct netdev_queue *dev_queue;
1769 struct tc_htb_opt *hopt; 1788 struct tc_htb_opt *hopt;
1770 u64 rate64, ceil64; 1789 u64 rate64, ceil64;
1771 int warn = 0; 1790 int warn = 0;
1772 1791
1773 /* extract all subattrs from opt attr 1792 /* extract all subattrs from opt attr */
1774 if (!opt) 1793 if (!opt)
1775 goto failure; 1794 goto failure;
1776 1795
1777 err = nla_parse_nested_deprecated(tb, 1796 err = nla_parse_nested_deprecated(tb, TCA_HTB_MAX, opt, htb_policy,
1778 ext !! 1797 NULL);
1779 if (err < 0) 1798 if (err < 0)
1780 goto failure; 1799 goto failure;
1781 1800
1782 err = -EINVAL; 1801 err = -EINVAL;
1783 if (tb[TCA_HTB_PARMS] == NULL) 1802 if (tb[TCA_HTB_PARMS] == NULL)
1784 goto failure; 1803 goto failure;
1785 1804
1786 parent = parentid == TC_H_ROOT ? NULL 1805 parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch);
1787 1806
1788 hopt = nla_data(tb[TCA_HTB_PARMS]); 1807 hopt = nla_data(tb[TCA_HTB_PARMS]);
1789 if (!hopt->rate.rate || !hopt->ceil.r 1808 if (!hopt->rate.rate || !hopt->ceil.rate)
1790 goto failure; 1809 goto failure;
1791 1810
1792 if (q->offload) { 1811 if (q->offload) {
1793 /* Options not supported by t 1812 /* Options not supported by the offload. */
1794 if (hopt->rate.overhead || ho 1813 if (hopt->rate.overhead || hopt->ceil.overhead) {
1795 NL_SET_ERR_MSG(extack 1814 NL_SET_ERR_MSG(extack, "HTB offload doesn't support the overhead parameter");
1796 goto failure; 1815 goto failure;
1797 } 1816 }
1798 if (hopt->rate.mpu || hopt->c 1817 if (hopt->rate.mpu || hopt->ceil.mpu) {
1799 NL_SET_ERR_MSG(extack 1818 NL_SET_ERR_MSG(extack, "HTB offload doesn't support the mpu parameter");
1800 goto failure; 1819 goto failure;
1801 } 1820 }
>> 1821 if (hopt->quantum) {
>> 1822 NL_SET_ERR_MSG(extack, "HTB offload doesn't support the quantum parameter");
>> 1823 goto failure;
>> 1824 }
>> 1825 if (hopt->prio) {
>> 1826 NL_SET_ERR_MSG(extack, "HTB offload doesn't support the prio parameter");
>> 1827 goto failure;
>> 1828 }
1802 } 1829 }
1803 1830
1804 /* Keeping backward compatible with r 1831 /* Keeping backward compatible with rate_table based iproute2 tc */
1805 if (hopt->rate.linklayer == TC_LINKLA 1832 if (hopt->rate.linklayer == TC_LINKLAYER_UNAWARE)
1806 qdisc_put_rtab(qdisc_get_rtab 1833 qdisc_put_rtab(qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB],
1807 1834 NULL));
1808 1835
1809 if (hopt->ceil.linklayer == TC_LINKLA 1836 if (hopt->ceil.linklayer == TC_LINKLAYER_UNAWARE)
1810 qdisc_put_rtab(qdisc_get_rtab 1837 qdisc_put_rtab(qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB],
1811 1838 NULL));
1812 1839
1813 rate64 = tb[TCA_HTB_RATE64] ? nla_get 1840 rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0;
1814 ceil64 = tb[TCA_HTB_CEIL64] ? nla_get 1841 ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0;
1815 1842
1816 if (!cl) { /* new class 1843 if (!cl) { /* new class */
1817 struct net_device *dev = qdis 1844 struct net_device *dev = qdisc_dev(sch);
1818 struct Qdisc *new_q, *old_q; 1845 struct Qdisc *new_q, *old_q;
1819 int prio; 1846 int prio;
1820 struct { 1847 struct {
1821 struct nlattr 1848 struct nlattr nla;
1822 struct gnet_estimator 1849 struct gnet_estimator opt;
1823 } est = { 1850 } est = {
1824 .nla = { 1851 .nla = {
1825 .nla_len 1852 .nla_len = nla_attr_size(sizeof(est.opt)),
1826 .nla_type 1853 .nla_type = TCA_RATE,
1827 }, 1854 },
1828 .opt = { 1855 .opt = {
1829 /* 4s interva 1856 /* 4s interval, 16s averaging constant */
1830 .interval 1857 .interval = 2,
1831 .ewma_log 1858 .ewma_log = 2,
1832 }, 1859 },
1833 }; 1860 };
1834 1861
1835 /* check for valid classid */ 1862 /* check for valid classid */
1836 if (!classid || TC_H_MAJ(clas 1863 if (!classid || TC_H_MAJ(classid ^ sch->handle) ||
1837 htb_find(classid, sch)) 1864 htb_find(classid, sch))
1838 goto failure; 1865 goto failure;
1839 1866
1840 /* check maximal depth */ 1867 /* check maximal depth */
1841 if (parent && parent->parent 1868 if (parent && parent->parent && parent->parent->level < 2) {
1842 NL_SET_ERR_MSG_MOD(ex !! 1869 pr_err("htb: tree is too deep\n");
1843 goto failure; 1870 goto failure;
1844 } 1871 }
1845 err = -ENOBUFS; 1872 err = -ENOBUFS;
1846 cl = kzalloc(sizeof(*cl), GFP 1873 cl = kzalloc(sizeof(*cl), GFP_KERNEL);
1847 if (!cl) 1874 if (!cl)
1848 goto failure; 1875 goto failure;
1849 1876
1850 gnet_stats_basic_sync_init(&c 1877 gnet_stats_basic_sync_init(&cl->bstats);
1851 gnet_stats_basic_sync_init(&c 1878 gnet_stats_basic_sync_init(&cl->bstats_bias);
1852 1879
1853 err = tcf_block_get(&cl->bloc 1880 err = tcf_block_get(&cl->block, &cl->filter_list, sch, extack);
1854 if (err) { 1881 if (err) {
1855 kfree(cl); 1882 kfree(cl);
1856 goto failure; 1883 goto failure;
1857 } 1884 }
1858 if (htb_rate_est || tca[TCA_R 1885 if (htb_rate_est || tca[TCA_RATE]) {
1859 err = gen_new_estimat 1886 err = gen_new_estimator(&cl->bstats, NULL,
1860 1887 &cl->rate_est,
1861 1888 NULL,
1862 1889 true,
1863 1890 tca[TCA_RATE] ? : &est.nla);
1864 if (err) 1891 if (err)
1865 goto err_bloc 1892 goto err_block_put;
1866 } 1893 }
1867 1894
1868 cl->children = 0; 1895 cl->children = 0;
1869 RB_CLEAR_NODE(&cl->pq_node); 1896 RB_CLEAR_NODE(&cl->pq_node);
1870 1897
1871 for (prio = 0; prio < TC_HTB_ 1898 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++)
1872 RB_CLEAR_NODE(&cl->no 1899 RB_CLEAR_NODE(&cl->node[prio]);
1873 1900
1874 cl->common.classid = classid; 1901 cl->common.classid = classid;
1875 1902
1876 /* Make sure nothing interrup 1903 /* Make sure nothing interrupts us in between of two
1877 * ndo_setup_tc calls. 1904 * ndo_setup_tc calls.
1878 */ 1905 */
1879 ASSERT_RTNL(); 1906 ASSERT_RTNL();
1880 1907
1881 /* create leaf qdisc early be 1908 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1882 * so that can't be used insi 1909 * so that can't be used inside of sch_tree_lock
1883 * -- thanks to Karlis Peisen 1910 * -- thanks to Karlis Peisenieks
1884 */ 1911 */
1885 if (!q->offload) { 1912 if (!q->offload) {
1886 dev_queue = sch->dev_ 1913 dev_queue = sch->dev_queue;
1887 } else if (!(parent && !paren 1914 } else if (!(parent && !parent->level)) {
1888 /* Assign a dev_queue 1915 /* Assign a dev_queue to this classid. */
1889 offload_opt = (struct 1916 offload_opt = (struct tc_htb_qopt_offload) {
1890 .command = TC 1917 .command = TC_HTB_LEAF_ALLOC_QUEUE,
1891 .classid = cl 1918 .classid = cl->common.classid,
1892 .parent_class 1919 .parent_classid = parent ?
1893 TC_H_ 1920 TC_H_MIN(parent->common.classid) :
1894 TC_HT 1921 TC_HTB_CLASSID_ROOT,
1895 .rate = max_t 1922 .rate = max_t(u64, hopt->rate.rate, rate64),
1896 .ceil = max_t 1923 .ceil = max_t(u64, hopt->ceil.rate, ceil64),
1897 .prio = hopt- <<
1898 .quantum = ho <<
1899 .extack = ext 1924 .extack = extack,
1900 }; 1925 };
1901 err = htb_offload(dev 1926 err = htb_offload(dev, &offload_opt);
1902 if (err) { 1927 if (err) {
1903 NL_SET_ERR_MS !! 1928 pr_err("htb: TC_HTB_LEAF_ALLOC_QUEUE failed with err = %d\n",
1904 !! 1929 err);
1905 goto err_kill 1930 goto err_kill_estimator;
1906 } 1931 }
1907 dev_queue = netdev_ge 1932 dev_queue = netdev_get_tx_queue(dev, offload_opt.qid);
1908 } else { /* First child. */ 1933 } else { /* First child. */
1909 dev_queue = htb_offlo 1934 dev_queue = htb_offload_get_queue(parent);
1910 old_q = htb_graft_hel 1935 old_q = htb_graft_helper(dev_queue, NULL);
1911 WARN_ON(old_q != pare 1936 WARN_ON(old_q != parent->leaf.q);
1912 offload_opt = (struct 1937 offload_opt = (struct tc_htb_qopt_offload) {
1913 .command = TC 1938 .command = TC_HTB_LEAF_TO_INNER,
1914 .classid = cl 1939 .classid = cl->common.classid,
1915 .parent_class 1940 .parent_classid =
1916 TC_H_ 1941 TC_H_MIN(parent->common.classid),
1917 .rate = max_t 1942 .rate = max_t(u64, hopt->rate.rate, rate64),
1918 .ceil = max_t 1943 .ceil = max_t(u64, hopt->ceil.rate, ceil64),
1919 .prio = hopt- <<
1920 .quantum = ho <<
1921 .extack = ext 1944 .extack = extack,
1922 }; 1945 };
1923 err = htb_offload(dev 1946 err = htb_offload(dev, &offload_opt);
1924 if (err) { 1947 if (err) {
1925 NL_SET_ERR_MS !! 1948 pr_err("htb: TC_HTB_LEAF_TO_INNER failed with err = %d\n",
1926 !! 1949 err);
1927 htb_graft_hel 1950 htb_graft_helper(dev_queue, old_q);
1928 goto err_kill 1951 goto err_kill_estimator;
1929 } 1952 }
1930 _bstats_update(&paren 1953 _bstats_update(&parent->bstats_bias,
1931 u64_st 1954 u64_stats_read(&old_q->bstats.bytes),
1932 u64_st 1955 u64_stats_read(&old_q->bstats.packets));
1933 qdisc_put(old_q); 1956 qdisc_put(old_q);
1934 } 1957 }
1935 new_q = qdisc_create_dflt(dev 1958 new_q = qdisc_create_dflt(dev_queue, &pfifo_qdisc_ops,
1936 cla 1959 classid, NULL);
1937 if (q->offload) { 1960 if (q->offload) {
1938 /* One ref for cl->le !! 1961 if (new_q) {
1939 if (new_q) !! 1962 htb_set_lockdep_class_child(new_q);
>> 1963 /* One ref for cl->leaf.q, the other for
>> 1964 * dev_queue->qdisc.
>> 1965 */
1940 qdisc_refcoun 1966 qdisc_refcount_inc(new_q);
>> 1967 }
1941 old_q = htb_graft_hel 1968 old_q = htb_graft_helper(dev_queue, new_q);
1942 /* No qdisc_put neede 1969 /* No qdisc_put needed. */
1943 WARN_ON(!(old_q->flag 1970 WARN_ON(!(old_q->flags & TCQ_F_BUILTIN));
1944 } 1971 }
1945 sch_tree_lock(sch); 1972 sch_tree_lock(sch);
1946 if (parent && !parent->level) 1973 if (parent && !parent->level) {
1947 /* turn parent into i 1974 /* turn parent into inner node */
1948 qdisc_purge_queue(par 1975 qdisc_purge_queue(parent->leaf.q);
1949 parent_qdisc = parent 1976 parent_qdisc = parent->leaf.q;
1950 if (parent->prio_acti 1977 if (parent->prio_activity)
1951 htb_deactivat 1978 htb_deactivate(q, parent);
1952 1979
1953 /* remove from evt li 1980 /* remove from evt list because of level change */
1954 if (parent->cmode != 1981 if (parent->cmode != HTB_CAN_SEND) {
1955 htb_safe_rb_e 1982 htb_safe_rb_erase(&parent->pq_node, &q->hlevel[0].wait_pq);
1956 parent->cmode 1983 parent->cmode = HTB_CAN_SEND;
1957 } 1984 }
1958 parent->level = (pare 1985 parent->level = (parent->parent ? parent->parent->level
1959 : TC 1986 : TC_HTB_MAXDEPTH) - 1;
1960 memset(&parent->inner 1987 memset(&parent->inner, 0, sizeof(parent->inner));
1961 } 1988 }
1962 1989
1963 /* leaf (we) needs elementary 1990 /* leaf (we) needs elementary qdisc */
1964 cl->leaf.q = new_q ? new_q : 1991 cl->leaf.q = new_q ? new_q : &noop_qdisc;
1965 if (q->offload) 1992 if (q->offload)
1966 cl->leaf.offload_queu 1993 cl->leaf.offload_queue = dev_queue;
1967 1994
1968 cl->parent = parent; 1995 cl->parent = parent;
1969 1996
1970 /* set class to be in HTB_CAN 1997 /* set class to be in HTB_CAN_SEND state */
1971 cl->tokens = PSCHED_TICKS2NS( 1998 cl->tokens = PSCHED_TICKS2NS(hopt->buffer);
1972 cl->ctokens = PSCHED_TICKS2NS 1999 cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer);
1973 cl->mbuffer = 60ULL * NSEC_PE 2000 cl->mbuffer = 60ULL * NSEC_PER_SEC; /* 1min */
1974 cl->t_c = ktime_get_ns(); 2001 cl->t_c = ktime_get_ns();
1975 cl->cmode = HTB_CAN_SEND; 2002 cl->cmode = HTB_CAN_SEND;
1976 2003
1977 /* attach to the hash list an 2004 /* attach to the hash list and parent's family */
1978 qdisc_class_hash_insert(&q->c 2005 qdisc_class_hash_insert(&q->clhash, &cl->common);
1979 if (parent) 2006 if (parent)
1980 parent->children++; 2007 parent->children++;
1981 if (cl->leaf.q != &noop_qdisc 2008 if (cl->leaf.q != &noop_qdisc)
1982 qdisc_hash_add(cl->le 2009 qdisc_hash_add(cl->leaf.q, true);
1983 } else { 2010 } else {
1984 if (tca[TCA_RATE]) { 2011 if (tca[TCA_RATE]) {
1985 err = gen_replace_est 2012 err = gen_replace_estimator(&cl->bstats, NULL,
1986 2013 &cl->rate_est,
1987 2014 NULL,
1988 2015 true,
1989 2016 tca[TCA_RATE]);
1990 if (err) 2017 if (err)
1991 return err; 2018 return err;
1992 } 2019 }
1993 2020
1994 if (q->offload) { 2021 if (q->offload) {
1995 struct net_device *de 2022 struct net_device *dev = qdisc_dev(sch);
1996 2023
1997 offload_opt = (struct 2024 offload_opt = (struct tc_htb_qopt_offload) {
1998 .command = TC 2025 .command = TC_HTB_NODE_MODIFY,
1999 .classid = cl 2026 .classid = cl->common.classid,
2000 .rate = max_t 2027 .rate = max_t(u64, hopt->rate.rate, rate64),
2001 .ceil = max_t 2028 .ceil = max_t(u64, hopt->ceil.rate, ceil64),
2002 .prio = hopt- <<
2003 .quantum = ho <<
2004 .extack = ext 2029 .extack = extack,
2005 }; 2030 };
2006 err = htb_offload(dev 2031 err = htb_offload(dev, &offload_opt);
2007 if (err) 2032 if (err)
2008 /* Estimator 2033 /* Estimator was replaced, and rollback may fail
2009 * as well, s 2034 * as well, so we don't try to recover it, and
2010 * the estima 2035 * the estimator won't work property with the
2011 * offload an 2036 * offload anyway, because bstats are updated
2012 * only when 2037 * only when the stats are queried.
2013 */ 2038 */
2014 return err; 2039 return err;
2015 } 2040 }
2016 2041
2017 sch_tree_lock(sch); 2042 sch_tree_lock(sch);
2018 } 2043 }
2019 2044
2020 psched_ratecfg_precompute(&cl->rate, 2045 psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64);
2021 psched_ratecfg_precompute(&cl->ceil, 2046 psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64);
2022 2047
2023 /* it used to be a nasty bug here, we 2048 /* it used to be a nasty bug here, we have to check that node
2024 * is really leaf before changing cl- 2049 * is really leaf before changing cl->leaf !
2025 */ 2050 */
2026 if (!cl->level) { 2051 if (!cl->level) {
2027 u64 quantum = cl->rate.rate_b 2052 u64 quantum = cl->rate.rate_bytes_ps;
2028 2053
2029 do_div(quantum, q->rate2quant 2054 do_div(quantum, q->rate2quantum);
2030 cl->quantum = min_t(u64, quan 2055 cl->quantum = min_t(u64, quantum, INT_MAX);
2031 2056
2032 if (!hopt->quantum && cl->qua 2057 if (!hopt->quantum && cl->quantum < 1000) {
2033 warn = -1; 2058 warn = -1;
2034 cl->quantum = 1000; 2059 cl->quantum = 1000;
2035 } 2060 }
2036 if (!hopt->quantum && cl->qua 2061 if (!hopt->quantum && cl->quantum > 200000) {
2037 warn = 1; 2062 warn = 1;
2038 cl->quantum = 200000; 2063 cl->quantum = 200000;
2039 } 2064 }
2040 if (hopt->quantum) 2065 if (hopt->quantum)
2041 cl->quantum = hopt->q 2066 cl->quantum = hopt->quantum;
2042 if ((cl->prio = hopt->prio) > 2067 if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO)
2043 cl->prio = TC_HTB_NUM 2068 cl->prio = TC_HTB_NUMPRIO - 1;
2044 } 2069 }
2045 2070
2046 cl->buffer = PSCHED_TICKS2NS(hopt->bu 2071 cl->buffer = PSCHED_TICKS2NS(hopt->buffer);
2047 cl->cbuffer = PSCHED_TICKS2NS(hopt->c 2072 cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer);
2048 2073
2049 sch_tree_unlock(sch); 2074 sch_tree_unlock(sch);
2050 qdisc_put(parent_qdisc); 2075 qdisc_put(parent_qdisc);
2051 2076
2052 if (warn) 2077 if (warn)
2053 NL_SET_ERR_MSG_FMT_MOD(extack !! 2078 pr_warn("HTB: quantum of class %X is %s. Consider r2q change.\n",
2054 "quant !! 2079 cl->common.classid, (warn == -1 ? "small" : "big"));
2055 cl->co <<
2056 2080
2057 qdisc_class_hash_grow(sch, &q->clhash 2081 qdisc_class_hash_grow(sch, &q->clhash);
2058 2082
2059 *arg = (unsigned long)cl; 2083 *arg = (unsigned long)cl;
2060 return 0; 2084 return 0;
2061 2085
2062 err_kill_estimator: 2086 err_kill_estimator:
2063 gen_kill_estimator(&cl->rate_est); 2087 gen_kill_estimator(&cl->rate_est);
2064 err_block_put: 2088 err_block_put:
2065 tcf_block_put(cl->block); 2089 tcf_block_put(cl->block);
2066 kfree(cl); 2090 kfree(cl);
2067 failure: 2091 failure:
2068 return err; 2092 return err;
2069 } 2093 }
2070 2094
2071 static struct tcf_block *htb_tcf_block(struct 2095 static struct tcf_block *htb_tcf_block(struct Qdisc *sch, unsigned long arg,
2072 struct 2096 struct netlink_ext_ack *extack)
2073 { 2097 {
2074 struct htb_sched *q = qdisc_priv(sch) 2098 struct htb_sched *q = qdisc_priv(sch);
2075 struct htb_class *cl = (struct htb_cl 2099 struct htb_class *cl = (struct htb_class *)arg;
2076 2100
2077 return cl ? cl->block : q->block; 2101 return cl ? cl->block : q->block;
2078 } 2102 }
2079 2103
2080 static unsigned long htb_bind_filter(struct Q 2104 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
2081 u32 clas 2105 u32 classid)
2082 { 2106 {
2083 struct htb_class *cl = htb_find(class 2107 struct htb_class *cl = htb_find(classid, sch);
2084 2108
2085 /*if (cl && !cl->level) return 0; 2109 /*if (cl && !cl->level) return 0;
2086 * The line above used to be there to 2110 * The line above used to be there to prevent attaching filters to
2087 * leaves. But at least tc_index filt 2111 * leaves. But at least tc_index filter uses this just to get class
2088 * for other reasons so that we have 2112 * for other reasons so that we have to allow for it.
2089 * ---- 2113 * ----
2090 * 19.6.2002 As Werner explained it i 2114 * 19.6.2002 As Werner explained it is ok - bind filter is just
2091 * another way to "lock" the class - 2115 * another way to "lock" the class - unlike "get" this lock can
2092 * be broken by class during destroy 2116 * be broken by class during destroy IIUC.
2093 */ 2117 */
2094 if (cl) 2118 if (cl)
2095 qdisc_class_get(&cl->common); !! 2119 cl->filter_cnt++;
2096 return (unsigned long)cl; 2120 return (unsigned long)cl;
2097 } 2121 }
2098 2122
2099 static void htb_unbind_filter(struct Qdisc *s 2123 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
2100 { 2124 {
2101 struct htb_class *cl = (struct htb_cl 2125 struct htb_class *cl = (struct htb_class *)arg;
2102 2126
2103 qdisc_class_put(&cl->common); !! 2127 if (cl)
>> 2128 cl->filter_cnt--;
2104 } 2129 }
2105 2130
2106 static void htb_walk(struct Qdisc *sch, struc 2131 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
2107 { 2132 {
2108 struct htb_sched *q = qdisc_priv(sch) 2133 struct htb_sched *q = qdisc_priv(sch);
2109 struct htb_class *cl; 2134 struct htb_class *cl;
2110 unsigned int i; 2135 unsigned int i;
2111 2136
2112 if (arg->stop) 2137 if (arg->stop)
2113 return; 2138 return;
2114 2139
2115 for (i = 0; i < q->clhash.hashsize; i 2140 for (i = 0; i < q->clhash.hashsize; i++) {
2116 hlist_for_each_entry(cl, &q-> 2141 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
2117 if (!tc_qdisc_stats_d !! 2142 if (arg->count < arg->skip) {
>> 2143 arg->count++;
>> 2144 continue;
>> 2145 }
>> 2146 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
>> 2147 arg->stop = 1;
2118 return; 2148 return;
>> 2149 }
>> 2150 arg->count++;
2119 } 2151 }
2120 } 2152 }
2121 } 2153 }
2122 2154
2123 static const struct Qdisc_class_ops htb_class 2155 static const struct Qdisc_class_ops htb_class_ops = {
2124 .select_queue = htb_select_qu 2156 .select_queue = htb_select_queue,
2125 .graft = htb_graft, 2157 .graft = htb_graft,
2126 .leaf = htb_leaf, 2158 .leaf = htb_leaf,
2127 .qlen_notify = htb_qlen_noti 2159 .qlen_notify = htb_qlen_notify,
2128 .find = htb_search, 2160 .find = htb_search,
2129 .change = htb_change_cl 2161 .change = htb_change_class,
2130 .delete = htb_delete, 2162 .delete = htb_delete,
2131 .walk = htb_walk, 2163 .walk = htb_walk,
2132 .tcf_block = htb_tcf_block 2164 .tcf_block = htb_tcf_block,
2133 .bind_tcf = htb_bind_filt 2165 .bind_tcf = htb_bind_filter,
2134 .unbind_tcf = htb_unbind_fi 2166 .unbind_tcf = htb_unbind_filter,
2135 .dump = htb_dump_clas 2167 .dump = htb_dump_class,
2136 .dump_stats = htb_dump_clas 2168 .dump_stats = htb_dump_class_stats,
2137 }; 2169 };
2138 2170
2139 static struct Qdisc_ops htb_qdisc_ops __read_ 2171 static struct Qdisc_ops htb_qdisc_ops __read_mostly = {
2140 .cl_ops = &htb_class_op 2172 .cl_ops = &htb_class_ops,
2141 .id = "htb", 2173 .id = "htb",
2142 .priv_size = sizeof(struct 2174 .priv_size = sizeof(struct htb_sched),
2143 .enqueue = htb_enqueue, 2175 .enqueue = htb_enqueue,
2144 .dequeue = htb_dequeue, 2176 .dequeue = htb_dequeue,
2145 .peek = qdisc_peek_de 2177 .peek = qdisc_peek_dequeued,
2146 .init = htb_init, 2178 .init = htb_init,
2147 .attach = htb_attach, 2179 .attach = htb_attach,
2148 .reset = htb_reset, 2180 .reset = htb_reset,
2149 .destroy = htb_destroy, 2181 .destroy = htb_destroy,
2150 .dump = htb_dump, 2182 .dump = htb_dump,
2151 .owner = THIS_MODULE, 2183 .owner = THIS_MODULE,
2152 }; 2184 };
2153 MODULE_ALIAS_NET_SCH("htb"); <<
2154 2185
2155 static int __init htb_module_init(void) 2186 static int __init htb_module_init(void)
2156 { 2187 {
2157 return register_qdisc(&htb_qdisc_ops) 2188 return register_qdisc(&htb_qdisc_ops);
2158 } 2189 }
2159 static void __exit htb_module_exit(void) 2190 static void __exit htb_module_exit(void)
2160 { 2191 {
2161 unregister_qdisc(&htb_qdisc_ops); 2192 unregister_qdisc(&htb_qdisc_ops);
2162 } 2193 }
2163 2194
2164 module_init(htb_module_init) 2195 module_init(htb_module_init)
2165 module_exit(htb_module_exit) 2196 module_exit(htb_module_exit)
2166 MODULE_LICENSE("GPL"); 2197 MODULE_LICENSE("GPL");
2167 MODULE_DESCRIPTION("Hierarchical Token Bucket <<
2168 2198
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.