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_packed bstats;
116 struct gnet_stats_basic_sync bstats_bi !! 117 struct gnet_stats_basic_packed 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, 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);
>> 1011 sch->q.qlen = 0;
>> 1012 sch->qstats.backlog = 0;
1017 memset(q->hlevel, 0, sizeof(q->hlevel 1013 memset(q->hlevel, 0, sizeof(q->hlevel));
1018 memset(q->row_mask, 0, sizeof(q->row_ 1014 memset(q->row_mask, 0, sizeof(q->row_mask));
1019 } 1015 }
1020 1016
1021 static const struct nla_policy htb_policy[TCA 1017 static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = {
1022 [TCA_HTB_PARMS] = { .len = sizeof(str 1018 [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) },
1023 [TCA_HTB_INIT] = { .len = sizeof(str 1019 [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) },
1024 [TCA_HTB_CTAB] = { .type = NLA_BINAR 1020 [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1025 [TCA_HTB_RTAB] = { .type = NLA_BINAR 1021 [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1026 [TCA_HTB_DIRECT_QLEN] = { .type = NLA 1022 [TCA_HTB_DIRECT_QLEN] = { .type = NLA_U32 },
1027 [TCA_HTB_RATE64] = { .type = NLA_U64 1023 [TCA_HTB_RATE64] = { .type = NLA_U64 },
1028 [TCA_HTB_CEIL64] = { .type = NLA_U64 1024 [TCA_HTB_CEIL64] = { .type = NLA_U64 },
1029 [TCA_HTB_OFFLOAD] = { .type = NLA_FLA 1025 [TCA_HTB_OFFLOAD] = { .type = NLA_FLAG },
1030 }; 1026 };
1031 1027
1032 static void htb_work_func(struct work_struct 1028 static void htb_work_func(struct work_struct *work)
1033 { 1029 {
1034 struct htb_sched *q = container_of(wo 1030 struct htb_sched *q = container_of(work, struct htb_sched, work);
1035 struct Qdisc *sch = q->watchdog.qdisc 1031 struct Qdisc *sch = q->watchdog.qdisc;
1036 1032
1037 rcu_read_lock(); 1033 rcu_read_lock();
1038 __netif_schedule(qdisc_root(sch)); 1034 __netif_schedule(qdisc_root(sch));
1039 rcu_read_unlock(); 1035 rcu_read_unlock();
1040 } 1036 }
1041 1037
>> 1038 static void htb_set_lockdep_class_child(struct Qdisc *q)
>> 1039 {
>> 1040 static struct lock_class_key child_key;
>> 1041
>> 1042 lockdep_set_class(qdisc_lock(q), &child_key);
>> 1043 }
>> 1044
1042 static int htb_offload(struct net_device *dev 1045 static int htb_offload(struct net_device *dev, struct tc_htb_qopt_offload *opt)
1043 { 1046 {
1044 return dev->netdev_ops->ndo_setup_tc( 1047 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_HTB, opt);
1045 } 1048 }
1046 1049
1047 static int htb_init(struct Qdisc *sch, struct 1050 static int htb_init(struct Qdisc *sch, struct nlattr *opt,
1048 struct netlink_ext_ack *e 1051 struct netlink_ext_ack *extack)
1049 { 1052 {
1050 struct net_device *dev = qdisc_dev(sc 1053 struct net_device *dev = qdisc_dev(sch);
1051 struct tc_htb_qopt_offload offload_op 1054 struct tc_htb_qopt_offload offload_opt;
1052 struct htb_sched *q = qdisc_priv(sch) 1055 struct htb_sched *q = qdisc_priv(sch);
1053 struct nlattr *tb[TCA_HTB_MAX + 1]; 1056 struct nlattr *tb[TCA_HTB_MAX + 1];
1054 struct tc_htb_glob *gopt; 1057 struct tc_htb_glob *gopt;
1055 unsigned int ntx; 1058 unsigned int ntx;
1056 bool offload; 1059 bool offload;
1057 int err; 1060 int err;
1058 1061
1059 qdisc_watchdog_init(&q->watchdog, sch 1062 qdisc_watchdog_init(&q->watchdog, sch);
1060 INIT_WORK(&q->work, htb_work_func); 1063 INIT_WORK(&q->work, htb_work_func);
1061 1064
1062 if (!opt) 1065 if (!opt)
1063 return -EINVAL; 1066 return -EINVAL;
1064 1067
1065 err = tcf_block_get(&q->block, &q->fi 1068 err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
1066 if (err) 1069 if (err)
1067 return err; 1070 return err;
1068 1071
1069 err = nla_parse_nested_deprecated(tb, 1072 err = nla_parse_nested_deprecated(tb, TCA_HTB_MAX, opt, htb_policy,
1070 NUL 1073 NULL);
1071 if (err < 0) 1074 if (err < 0)
1072 return err; 1075 return err;
1073 1076
1074 if (!tb[TCA_HTB_INIT]) 1077 if (!tb[TCA_HTB_INIT])
1075 return -EINVAL; 1078 return -EINVAL;
1076 1079
1077 gopt = nla_data(tb[TCA_HTB_INIT]); 1080 gopt = nla_data(tb[TCA_HTB_INIT]);
1078 if (gopt->version != HTB_VER >> 16) 1081 if (gopt->version != HTB_VER >> 16)
1079 return -EINVAL; 1082 return -EINVAL;
1080 1083
1081 offload = nla_get_flag(tb[TCA_HTB_OFF 1084 offload = nla_get_flag(tb[TCA_HTB_OFFLOAD]);
1082 1085
1083 if (offload) { 1086 if (offload) {
1084 if (sch->parent != TC_H_ROOT) !! 1087 if (sch->parent != TC_H_ROOT)
1085 NL_SET_ERR_MSG(extack <<
1086 return -EOPNOTSUPP; 1088 return -EOPNOTSUPP;
1087 } <<
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 return -EOPNOTSUPP; 1091 return -EOPNOTSUPP;
1092 } <<
1093 1092
1094 q->num_direct_qdiscs = dev->r 1093 q->num_direct_qdiscs = dev->real_num_tx_queues;
1095 q->direct_qdiscs = kcalloc(q- 1094 q->direct_qdiscs = kcalloc(q->num_direct_qdiscs,
1096 si 1095 sizeof(*q->direct_qdiscs),
1097 GF 1096 GFP_KERNEL);
1098 if (!q->direct_qdiscs) 1097 if (!q->direct_qdiscs)
1099 return -ENOMEM; 1098 return -ENOMEM;
1100 } 1099 }
1101 1100
1102 err = qdisc_class_hash_init(&q->clhas 1101 err = qdisc_class_hash_init(&q->clhash);
1103 if (err < 0) 1102 if (err < 0)
1104 return err; !! 1103 goto err_free_direct_qdiscs;
>> 1104
>> 1105 qdisc_skb_head_init(&q->direct_queue);
1105 1106
1106 if (tb[TCA_HTB_DIRECT_QLEN]) 1107 if (tb[TCA_HTB_DIRECT_QLEN])
1107 q->direct_qlen = nla_get_u32( 1108 q->direct_qlen = nla_get_u32(tb[TCA_HTB_DIRECT_QLEN]);
1108 else 1109 else
1109 q->direct_qlen = qdisc_dev(sc 1110 q->direct_qlen = qdisc_dev(sch)->tx_queue_len;
1110 1111
1111 if ((q->rate2quantum = gopt->rate2qua 1112 if ((q->rate2quantum = gopt->rate2quantum) < 1)
1112 q->rate2quantum = 1; 1113 q->rate2quantum = 1;
1113 q->defcls = gopt->defcls; 1114 q->defcls = gopt->defcls;
1114 1115
1115 if (!offload) 1116 if (!offload)
1116 return 0; 1117 return 0;
1117 1118
1118 for (ntx = 0; ntx < q->num_direct_qdi 1119 for (ntx = 0; ntx < q->num_direct_qdiscs; ntx++) {
1119 struct netdev_queue *dev_queu 1120 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, ntx);
1120 struct Qdisc *qdisc; 1121 struct Qdisc *qdisc;
1121 1122
1122 qdisc = qdisc_create_dflt(dev 1123 qdisc = qdisc_create_dflt(dev_queue, &pfifo_qdisc_ops,
1123 TC_ 1124 TC_H_MAKE(sch->handle, 0), extack);
1124 if (!qdisc) { 1125 if (!qdisc) {
1125 return -ENOMEM; !! 1126 err = -ENOMEM;
>> 1127 goto err_free_qdiscs;
1126 } 1128 }
1127 1129
>> 1130 htb_set_lockdep_class_child(qdisc);
1128 q->direct_qdiscs[ntx] = qdisc 1131 q->direct_qdiscs[ntx] = qdisc;
1129 qdisc->flags |= TCQ_F_ONETXQU 1132 qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1130 } 1133 }
1131 1134
1132 sch->flags |= TCQ_F_MQROOT; 1135 sch->flags |= TCQ_F_MQROOT;
1133 1136
1134 offload_opt = (struct tc_htb_qopt_off 1137 offload_opt = (struct tc_htb_qopt_offload) {
1135 .command = TC_HTB_CREATE, 1138 .command = TC_HTB_CREATE,
1136 .parent_classid = TC_H_MAJ(sc 1139 .parent_classid = TC_H_MAJ(sch->handle) >> 16,
1137 .classid = TC_H_MIN(q->defcls 1140 .classid = TC_H_MIN(q->defcls),
1138 .extack = extack, 1141 .extack = extack,
1139 }; 1142 };
1140 err = htb_offload(dev, &offload_opt); 1143 err = htb_offload(dev, &offload_opt);
1141 if (err) 1144 if (err)
1142 return err; !! 1145 goto err_free_qdiscs;
1143 1146
1144 /* Defer this assignment, so that htb 1147 /* Defer this assignment, so that htb_destroy skips offload-related
1145 * parts (especially calling ndo_setu 1148 * parts (especially calling ndo_setup_tc) on errors.
1146 */ 1149 */
1147 q->offload = true; 1150 q->offload = true;
1148 1151
1149 return 0; 1152 return 0;
>> 1153
>> 1154 err_free_qdiscs:
>> 1155 for (ntx = 0; ntx < q->num_direct_qdiscs && q->direct_qdiscs[ntx];
>> 1156 ntx++)
>> 1157 qdisc_put(q->direct_qdiscs[ntx]);
>> 1158
>> 1159 qdisc_class_hash_destroy(&q->clhash);
>> 1160 /* Prevent use-after-free and double-free when htb_destroy gets called.
>> 1161 */
>> 1162 q->clhash.hash = NULL;
>> 1163 q->clhash.hashsize = 0;
>> 1164
>> 1165 err_free_direct_qdiscs:
>> 1166 kfree(q->direct_qdiscs);
>> 1167 q->direct_qdiscs = NULL;
>> 1168 return err;
1150 } 1169 }
1151 1170
1152 static void htb_attach_offload(struct Qdisc * 1171 static void htb_attach_offload(struct Qdisc *sch)
1153 { 1172 {
1154 struct net_device *dev = qdisc_dev(sc 1173 struct net_device *dev = qdisc_dev(sch);
1155 struct htb_sched *q = qdisc_priv(sch) 1174 struct htb_sched *q = qdisc_priv(sch);
1156 unsigned int ntx; 1175 unsigned int ntx;
1157 1176
1158 for (ntx = 0; ntx < q->num_direct_qdi 1177 for (ntx = 0; ntx < q->num_direct_qdiscs; ntx++) {
1159 struct Qdisc *old, *qdisc = q 1178 struct Qdisc *old, *qdisc = q->direct_qdiscs[ntx];
1160 1179
1161 old = dev_graft_qdisc(qdisc-> 1180 old = dev_graft_qdisc(qdisc->dev_queue, qdisc);
1162 qdisc_put(old); 1181 qdisc_put(old);
1163 qdisc_hash_add(qdisc, false); 1182 qdisc_hash_add(qdisc, false);
1164 } 1183 }
1165 for (ntx = q->num_direct_qdiscs; ntx 1184 for (ntx = q->num_direct_qdiscs; ntx < dev->num_tx_queues; ntx++) {
1166 struct netdev_queue *dev_queu 1185 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, ntx);
1167 struct Qdisc *old = dev_graft 1186 struct Qdisc *old = dev_graft_qdisc(dev_queue, NULL);
1168 1187
1169 qdisc_put(old); 1188 qdisc_put(old);
1170 } 1189 }
1171 1190
1172 kfree(q->direct_qdiscs); 1191 kfree(q->direct_qdiscs);
1173 q->direct_qdiscs = NULL; 1192 q->direct_qdiscs = NULL;
1174 } 1193 }
1175 1194
1176 static void htb_attach_software(struct Qdisc 1195 static void htb_attach_software(struct Qdisc *sch)
1177 { 1196 {
1178 struct net_device *dev = qdisc_dev(sc 1197 struct net_device *dev = qdisc_dev(sch);
1179 unsigned int ntx; 1198 unsigned int ntx;
1180 1199
1181 /* Resemble qdisc_graft behavior. */ 1200 /* Resemble qdisc_graft behavior. */
1182 for (ntx = 0; ntx < dev->num_tx_queue 1201 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
1183 struct netdev_queue *dev_queu 1202 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, ntx);
1184 struct Qdisc *old = dev_graft 1203 struct Qdisc *old = dev_graft_qdisc(dev_queue, sch);
1185 1204
1186 qdisc_refcount_inc(sch); 1205 qdisc_refcount_inc(sch);
1187 1206
1188 qdisc_put(old); 1207 qdisc_put(old);
1189 } 1208 }
1190 } 1209 }
1191 1210
1192 static void htb_attach(struct Qdisc *sch) 1211 static void htb_attach(struct Qdisc *sch)
1193 { 1212 {
1194 struct htb_sched *q = qdisc_priv(sch) 1213 struct htb_sched *q = qdisc_priv(sch);
1195 1214
1196 if (q->offload) 1215 if (q->offload)
1197 htb_attach_offload(sch); 1216 htb_attach_offload(sch);
1198 else 1217 else
1199 htb_attach_software(sch); 1218 htb_attach_software(sch);
1200 } 1219 }
1201 1220
1202 static int htb_dump(struct Qdisc *sch, struct 1221 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1203 { 1222 {
1204 struct htb_sched *q = qdisc_priv(sch) 1223 struct htb_sched *q = qdisc_priv(sch);
1205 struct nlattr *nest; 1224 struct nlattr *nest;
1206 struct tc_htb_glob gopt; 1225 struct tc_htb_glob gopt;
1207 1226
1208 if (q->offload) 1227 if (q->offload)
1209 sch->flags |= TCQ_F_OFFLOADED 1228 sch->flags |= TCQ_F_OFFLOADED;
1210 else 1229 else
1211 sch->flags &= ~TCQ_F_OFFLOADE 1230 sch->flags &= ~TCQ_F_OFFLOADED;
1212 1231
1213 sch->qstats.overlimits = q->overlimit 1232 sch->qstats.overlimits = q->overlimits;
1214 /* Its safe to not acquire qdisc lock 1233 /* Its safe to not acquire qdisc lock. As we hold RTNL,
1215 * no change can happen on the qdisc 1234 * no change can happen on the qdisc parameters.
1216 */ 1235 */
1217 1236
1218 gopt.direct_pkts = q->direct_pkts; 1237 gopt.direct_pkts = q->direct_pkts;
1219 gopt.version = HTB_VER; 1238 gopt.version = HTB_VER;
1220 gopt.rate2quantum = q->rate2quantum; 1239 gopt.rate2quantum = q->rate2quantum;
1221 gopt.defcls = q->defcls; 1240 gopt.defcls = q->defcls;
1222 gopt.debug = 0; 1241 gopt.debug = 0;
1223 1242
1224 nest = nla_nest_start_noflag(skb, TCA 1243 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1225 if (nest == NULL) 1244 if (nest == NULL)
1226 goto nla_put_failure; 1245 goto nla_put_failure;
1227 if (nla_put(skb, TCA_HTB_INIT, sizeof 1246 if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt) ||
1228 nla_put_u32(skb, TCA_HTB_DIRECT_Q 1247 nla_put_u32(skb, TCA_HTB_DIRECT_QLEN, q->direct_qlen))
1229 goto nla_put_failure; 1248 goto nla_put_failure;
1230 if (q->offload && nla_put_flag(skb, T 1249 if (q->offload && nla_put_flag(skb, TCA_HTB_OFFLOAD))
1231 goto nla_put_failure; 1250 goto nla_put_failure;
1232 1251
1233 return nla_nest_end(skb, nest); 1252 return nla_nest_end(skb, nest);
1234 1253
1235 nla_put_failure: 1254 nla_put_failure:
1236 nla_nest_cancel(skb, nest); 1255 nla_nest_cancel(skb, nest);
1237 return -1; 1256 return -1;
1238 } 1257 }
1239 1258
1240 static int htb_dump_class(struct Qdisc *sch, 1259 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1241 struct sk_buff *skb 1260 struct sk_buff *skb, struct tcmsg *tcm)
1242 { 1261 {
1243 struct htb_class *cl = (struct htb_cl 1262 struct htb_class *cl = (struct htb_class *)arg;
1244 struct htb_sched *q = qdisc_priv(sch) 1263 struct htb_sched *q = qdisc_priv(sch);
1245 struct nlattr *nest; 1264 struct nlattr *nest;
1246 struct tc_htb_opt opt; 1265 struct tc_htb_opt opt;
1247 1266
1248 /* Its safe to not acquire qdisc lock 1267 /* Its safe to not acquire qdisc lock. As we hold RTNL,
1249 * no change can happen on the class 1268 * no change can happen on the class parameters.
1250 */ 1269 */
1251 tcm->tcm_parent = cl->parent ? cl->pa 1270 tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT;
1252 tcm->tcm_handle = cl->common.classid; 1271 tcm->tcm_handle = cl->common.classid;
1253 if (!cl->level && cl->leaf.q) 1272 if (!cl->level && cl->leaf.q)
1254 tcm->tcm_info = cl->leaf.q->h 1273 tcm->tcm_info = cl->leaf.q->handle;
1255 1274
1256 nest = nla_nest_start_noflag(skb, TCA 1275 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1257 if (nest == NULL) 1276 if (nest == NULL)
1258 goto nla_put_failure; 1277 goto nla_put_failure;
1259 1278
1260 memset(&opt, 0, sizeof(opt)); 1279 memset(&opt, 0, sizeof(opt));
1261 1280
1262 psched_ratecfg_getrate(&opt.rate, &cl 1281 psched_ratecfg_getrate(&opt.rate, &cl->rate);
1263 opt.buffer = PSCHED_NS2TICKS(cl->buff 1282 opt.buffer = PSCHED_NS2TICKS(cl->buffer);
1264 psched_ratecfg_getrate(&opt.ceil, &cl 1283 psched_ratecfg_getrate(&opt.ceil, &cl->ceil);
1265 opt.cbuffer = PSCHED_NS2TICKS(cl->cbu 1284 opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer);
1266 opt.quantum = cl->quantum; 1285 opt.quantum = cl->quantum;
1267 opt.prio = cl->prio; 1286 opt.prio = cl->prio;
1268 opt.level = cl->level; 1287 opt.level = cl->level;
1269 if (nla_put(skb, TCA_HTB_PARMS, sizeo 1288 if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt))
1270 goto nla_put_failure; 1289 goto nla_put_failure;
1271 if (q->offload && nla_put_flag(skb, T 1290 if (q->offload && nla_put_flag(skb, TCA_HTB_OFFLOAD))
1272 goto nla_put_failure; 1291 goto nla_put_failure;
1273 if ((cl->rate.rate_bytes_ps >= (1ULL 1292 if ((cl->rate.rate_bytes_ps >= (1ULL << 32)) &&
1274 nla_put_u64_64bit(skb, TCA_HTB_RA 1293 nla_put_u64_64bit(skb, TCA_HTB_RATE64, cl->rate.rate_bytes_ps,
1275 TCA_HTB_PAD)) 1294 TCA_HTB_PAD))
1276 goto nla_put_failure; 1295 goto nla_put_failure;
1277 if ((cl->ceil.rate_bytes_ps >= (1ULL 1296 if ((cl->ceil.rate_bytes_ps >= (1ULL << 32)) &&
1278 nla_put_u64_64bit(skb, TCA_HTB_CE 1297 nla_put_u64_64bit(skb, TCA_HTB_CEIL64, cl->ceil.rate_bytes_ps,
1279 TCA_HTB_PAD)) 1298 TCA_HTB_PAD))
1280 goto nla_put_failure; 1299 goto nla_put_failure;
1281 1300
1282 return nla_nest_end(skb, nest); 1301 return nla_nest_end(skb, nest);
1283 1302
1284 nla_put_failure: 1303 nla_put_failure:
1285 nla_nest_cancel(skb, nest); 1304 nla_nest_cancel(skb, nest);
1286 return -1; 1305 return -1;
1287 } 1306 }
1288 1307
1289 static void htb_offload_aggregate_stats(struc 1308 static void htb_offload_aggregate_stats(struct htb_sched *q,
1290 struc 1309 struct htb_class *cl)
1291 { 1310 {
1292 u64 bytes = 0, packets = 0; <<
1293 struct htb_class *c; 1311 struct htb_class *c;
1294 unsigned int i; 1312 unsigned int i;
1295 1313
1296 gnet_stats_basic_sync_init(&cl->bstat !! 1314 memset(&cl->bstats, 0, sizeof(cl->bstats));
1297 1315
1298 for (i = 0; i < q->clhash.hashsize; i 1316 for (i = 0; i < q->clhash.hashsize; i++) {
1299 hlist_for_each_entry(c, &q->c 1317 hlist_for_each_entry(c, &q->clhash.hash[i], common.hnode) {
1300 struct htb_class *p = 1318 struct htb_class *p = c;
1301 1319
1302 while (p && p->level 1320 while (p && p->level < cl->level)
1303 p = p->parent 1321 p = p->parent;
1304 1322
1305 if (p != cl) 1323 if (p != cl)
1306 continue; 1324 continue;
1307 1325
1308 bytes += u64_stats_re !! 1326 cl->bstats.bytes += c->bstats_bias.bytes;
1309 packets += u64_stats_ !! 1327 cl->bstats.packets += c->bstats_bias.packets;
1310 if (c->level == 0) { 1328 if (c->level == 0) {
1311 bytes += u64_ !! 1329 cl->bstats.bytes += c->leaf.q->bstats.bytes;
1312 packets += u6 !! 1330 cl->bstats.packets += c->leaf.q->bstats.packets;
1313 } 1331 }
1314 } 1332 }
1315 } 1333 }
1316 _bstats_update(&cl->bstats, bytes, pa <<
1317 } 1334 }
1318 1335
1319 static int 1336 static int
1320 htb_dump_class_stats(struct Qdisc *sch, unsig 1337 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d)
1321 { 1338 {
1322 struct htb_class *cl = (struct htb_cl 1339 struct htb_class *cl = (struct htb_class *)arg;
1323 struct htb_sched *q = qdisc_priv(sch) 1340 struct htb_sched *q = qdisc_priv(sch);
1324 struct gnet_stats_queue qs = { 1341 struct gnet_stats_queue qs = {
1325 .drops = cl->drops, 1342 .drops = cl->drops,
1326 .overlimits = cl->overlimits, 1343 .overlimits = cl->overlimits,
1327 }; 1344 };
1328 __u32 qlen = 0; 1345 __u32 qlen = 0;
1329 1346
1330 if (!cl->level && cl->leaf.q) 1347 if (!cl->level && cl->leaf.q)
1331 qdisc_qstats_qlen_backlog(cl- 1348 qdisc_qstats_qlen_backlog(cl->leaf.q, &qlen, &qs.backlog);
1332 1349
1333 cl->xstats.tokens = clamp_t(s64, PSCH 1350 cl->xstats.tokens = clamp_t(s64, PSCHED_NS2TICKS(cl->tokens),
1334 INT_MIN, 1351 INT_MIN, INT_MAX);
1335 cl->xstats.ctokens = clamp_t(s64, PSC 1352 cl->xstats.ctokens = clamp_t(s64, PSCHED_NS2TICKS(cl->ctokens),
1336 INT_MIN, 1353 INT_MIN, INT_MAX);
1337 1354
1338 if (q->offload) { 1355 if (q->offload) {
1339 if (!cl->level) { 1356 if (!cl->level) {
1340 if (cl->leaf.q) 1357 if (cl->leaf.q)
1341 cl->bstats = 1358 cl->bstats = cl->leaf.q->bstats;
1342 else 1359 else
1343 gnet_stats_ba !! 1360 memset(&cl->bstats, 0, sizeof(cl->bstats));
1344 _bstats_update(&cl->b !! 1361 cl->bstats.bytes += cl->bstats_bias.bytes;
1345 u64_st !! 1362 cl->bstats.packets += cl->bstats_bias.packets;
1346 u64_st <<
1347 } else { 1363 } else {
1348 htb_offload_aggregate 1364 htb_offload_aggregate_stats(q, cl);
1349 } 1365 }
1350 } 1366 }
1351 1367
1352 if (gnet_stats_copy_basic(d, NULL, &c !! 1368 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
>> 1369 d, NULL, &cl->bstats) < 0 ||
1353 gnet_stats_copy_rate_est(d, &cl-> 1370 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1354 gnet_stats_copy_queue(d, NULL, &q 1371 gnet_stats_copy_queue(d, NULL, &qs, qlen) < 0)
1355 return -1; 1372 return -1;
1356 1373
1357 return gnet_stats_copy_app(d, &cl->xs 1374 return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1358 } 1375 }
1359 1376
1360 static struct netdev_queue * 1377 static struct netdev_queue *
1361 htb_select_queue(struct Qdisc *sch, struct tc 1378 htb_select_queue(struct Qdisc *sch, struct tcmsg *tcm)
1362 { 1379 {
1363 struct net_device *dev = qdisc_dev(sc 1380 struct net_device *dev = qdisc_dev(sch);
1364 struct tc_htb_qopt_offload offload_op 1381 struct tc_htb_qopt_offload offload_opt;
1365 struct htb_sched *q = qdisc_priv(sch) 1382 struct htb_sched *q = qdisc_priv(sch);
1366 int err; 1383 int err;
1367 1384
1368 if (!q->offload) 1385 if (!q->offload)
1369 return sch->dev_queue; 1386 return sch->dev_queue;
1370 1387
1371 offload_opt = (struct tc_htb_qopt_off 1388 offload_opt = (struct tc_htb_qopt_offload) {
1372 .command = TC_HTB_LEAF_QUERY_ 1389 .command = TC_HTB_LEAF_QUERY_QUEUE,
1373 .classid = TC_H_MIN(tcm->tcm_ 1390 .classid = TC_H_MIN(tcm->tcm_parent),
1374 }; 1391 };
1375 err = htb_offload(dev, &offload_opt); 1392 err = htb_offload(dev, &offload_opt);
1376 if (err || offload_opt.qid >= dev->nu 1393 if (err || offload_opt.qid >= dev->num_tx_queues)
1377 return NULL; 1394 return NULL;
1378 return netdev_get_tx_queue(dev, offlo 1395 return netdev_get_tx_queue(dev, offload_opt.qid);
1379 } 1396 }
1380 1397
1381 static struct Qdisc * 1398 static struct Qdisc *
1382 htb_graft_helper(struct netdev_queue *dev_que 1399 htb_graft_helper(struct netdev_queue *dev_queue, struct Qdisc *new_q)
1383 { 1400 {
1384 struct net_device *dev = dev_queue->d 1401 struct net_device *dev = dev_queue->dev;
1385 struct Qdisc *old_q; 1402 struct Qdisc *old_q;
1386 1403
1387 if (dev->flags & IFF_UP) 1404 if (dev->flags & IFF_UP)
1388 dev_deactivate(dev); 1405 dev_deactivate(dev);
1389 old_q = dev_graft_qdisc(dev_queue, ne 1406 old_q = dev_graft_qdisc(dev_queue, new_q);
1390 if (new_q) 1407 if (new_q)
1391 new_q->flags |= TCQ_F_ONETXQU 1408 new_q->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1392 if (dev->flags & IFF_UP) 1409 if (dev->flags & IFF_UP)
1393 dev_activate(dev); 1410 dev_activate(dev);
1394 1411
1395 return old_q; 1412 return old_q;
1396 } 1413 }
1397 1414
1398 static struct netdev_queue *htb_offload_get_q 1415 static struct netdev_queue *htb_offload_get_queue(struct htb_class *cl)
1399 { 1416 {
1400 struct netdev_queue *queue; 1417 struct netdev_queue *queue;
1401 1418
1402 queue = cl->leaf.offload_queue; 1419 queue = cl->leaf.offload_queue;
1403 if (!(cl->leaf.q->flags & TCQ_F_BUILT 1420 if (!(cl->leaf.q->flags & TCQ_F_BUILTIN))
1404 WARN_ON(cl->leaf.q->dev_queue 1421 WARN_ON(cl->leaf.q->dev_queue != queue);
1405 1422
1406 return queue; 1423 return queue;
1407 } 1424 }
1408 1425
1409 static void htb_offload_move_qdisc(struct Qdi 1426 static void htb_offload_move_qdisc(struct Qdisc *sch, struct htb_class *cl_old,
1410 struct htb 1427 struct htb_class *cl_new, bool destroying)
1411 { 1428 {
1412 struct netdev_queue *queue_old, *queu 1429 struct netdev_queue *queue_old, *queue_new;
1413 struct net_device *dev = qdisc_dev(sc 1430 struct net_device *dev = qdisc_dev(sch);
1414 1431
1415 queue_old = htb_offload_get_queue(cl_ 1432 queue_old = htb_offload_get_queue(cl_old);
1416 queue_new = htb_offload_get_queue(cl_ 1433 queue_new = htb_offload_get_queue(cl_new);
1417 1434
1418 if (!destroying) { 1435 if (!destroying) {
1419 struct Qdisc *qdisc; 1436 struct Qdisc *qdisc;
1420 1437
1421 if (dev->flags & IFF_UP) 1438 if (dev->flags & IFF_UP)
1422 dev_deactivate(dev); 1439 dev_deactivate(dev);
1423 qdisc = dev_graft_qdisc(queue 1440 qdisc = dev_graft_qdisc(queue_old, NULL);
1424 WARN_ON(qdisc != cl_old->leaf 1441 WARN_ON(qdisc != cl_old->leaf.q);
1425 } 1442 }
1426 1443
1427 if (!(cl_old->leaf.q->flags & TCQ_F_B 1444 if (!(cl_old->leaf.q->flags & TCQ_F_BUILTIN))
1428 cl_old->leaf.q->dev_queue = q 1445 cl_old->leaf.q->dev_queue = queue_new;
1429 cl_old->leaf.offload_queue = queue_ne 1446 cl_old->leaf.offload_queue = queue_new;
1430 1447
1431 if (!destroying) { 1448 if (!destroying) {
1432 struct Qdisc *qdisc; 1449 struct Qdisc *qdisc;
1433 1450
1434 qdisc = dev_graft_qdisc(queue 1451 qdisc = dev_graft_qdisc(queue_new, cl_old->leaf.q);
1435 if (dev->flags & IFF_UP) 1452 if (dev->flags & IFF_UP)
1436 dev_activate(dev); 1453 dev_activate(dev);
1437 WARN_ON(!(qdisc->flags & TCQ_ 1454 WARN_ON(!(qdisc->flags & TCQ_F_BUILTIN));
1438 } 1455 }
1439 } 1456 }
1440 1457
1441 static int htb_graft(struct Qdisc *sch, unsig 1458 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1442 struct Qdisc **old, stru 1459 struct Qdisc **old, struct netlink_ext_ack *extack)
1443 { 1460 {
1444 struct netdev_queue *dev_queue = sch- 1461 struct netdev_queue *dev_queue = sch->dev_queue;
1445 struct htb_class *cl = (struct htb_cl 1462 struct htb_class *cl = (struct htb_class *)arg;
1446 struct htb_sched *q = qdisc_priv(sch) 1463 struct htb_sched *q = qdisc_priv(sch);
1447 struct Qdisc *old_q; 1464 struct Qdisc *old_q;
1448 1465
1449 if (cl->level) 1466 if (cl->level)
1450 return -EINVAL; 1467 return -EINVAL;
1451 1468
1452 if (q->offload) 1469 if (q->offload)
1453 dev_queue = htb_offload_get_q 1470 dev_queue = htb_offload_get_queue(cl);
1454 1471
1455 if (!new) { 1472 if (!new) {
1456 new = qdisc_create_dflt(dev_q 1473 new = qdisc_create_dflt(dev_queue, &pfifo_qdisc_ops,
1457 cl->c 1474 cl->common.classid, extack);
1458 if (!new) 1475 if (!new)
1459 return -ENOBUFS; 1476 return -ENOBUFS;
1460 } 1477 }
1461 1478
1462 if (q->offload) { 1479 if (q->offload) {
>> 1480 htb_set_lockdep_class_child(new);
1463 /* One ref for cl->leaf.q, th 1481 /* One ref for cl->leaf.q, the other for dev_queue->qdisc. */
1464 qdisc_refcount_inc(new); 1482 qdisc_refcount_inc(new);
1465 old_q = htb_graft_helper(dev_ 1483 old_q = htb_graft_helper(dev_queue, new);
1466 } 1484 }
1467 1485
1468 *old = qdisc_replace(sch, new, &cl->l 1486 *old = qdisc_replace(sch, new, &cl->leaf.q);
1469 1487
1470 if (q->offload) { 1488 if (q->offload) {
1471 WARN_ON(old_q != *old); 1489 WARN_ON(old_q != *old);
1472 qdisc_put(old_q); 1490 qdisc_put(old_q);
1473 } 1491 }
1474 1492
1475 return 0; 1493 return 0;
1476 } 1494 }
1477 1495
1478 static struct Qdisc *htb_leaf(struct Qdisc *s 1496 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg)
1479 { 1497 {
1480 struct htb_class *cl = (struct htb_cl 1498 struct htb_class *cl = (struct htb_class *)arg;
1481 return !cl->level ? cl->leaf.q : NULL 1499 return !cl->level ? cl->leaf.q : NULL;
1482 } 1500 }
1483 1501
1484 static void htb_qlen_notify(struct Qdisc *sch 1502 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg)
1485 { 1503 {
1486 struct htb_class *cl = (struct htb_cl 1504 struct htb_class *cl = (struct htb_class *)arg;
1487 1505
1488 htb_deactivate(qdisc_priv(sch), cl); 1506 htb_deactivate(qdisc_priv(sch), cl);
1489 } 1507 }
1490 1508
1491 static inline int htb_parent_last_child(struc 1509 static inline int htb_parent_last_child(struct htb_class *cl)
1492 { 1510 {
1493 if (!cl->parent) 1511 if (!cl->parent)
1494 /* the root class */ 1512 /* the root class */
1495 return 0; 1513 return 0;
1496 if (cl->parent->children > 1) 1514 if (cl->parent->children > 1)
1497 /* not the last child */ 1515 /* not the last child */
1498 return 0; 1516 return 0;
1499 return 1; 1517 return 1;
1500 } 1518 }
1501 1519
1502 static void htb_parent_to_leaf(struct Qdisc * 1520 static void htb_parent_to_leaf(struct Qdisc *sch, struct htb_class *cl,
1503 struct Qdisc * 1521 struct Qdisc *new_q)
1504 { 1522 {
1505 struct htb_sched *q = qdisc_priv(sch) 1523 struct htb_sched *q = qdisc_priv(sch);
1506 struct htb_class *parent = cl->parent 1524 struct htb_class *parent = cl->parent;
1507 1525
1508 WARN_ON(cl->level || !cl->leaf.q || c 1526 WARN_ON(cl->level || !cl->leaf.q || cl->prio_activity);
1509 1527
1510 if (parent->cmode != HTB_CAN_SEND) 1528 if (parent->cmode != HTB_CAN_SEND)
1511 htb_safe_rb_erase(&parent->pq 1529 htb_safe_rb_erase(&parent->pq_node,
1512 &q->hlevel[ 1530 &q->hlevel[parent->level].wait_pq);
1513 1531
1514 parent->level = 0; 1532 parent->level = 0;
1515 memset(&parent->inner, 0, sizeof(pare 1533 memset(&parent->inner, 0, sizeof(parent->inner));
1516 parent->leaf.q = new_q ? new_q : &noo 1534 parent->leaf.q = new_q ? new_q : &noop_qdisc;
1517 parent->tokens = parent->buffer; 1535 parent->tokens = parent->buffer;
1518 parent->ctokens = parent->cbuffer; 1536 parent->ctokens = parent->cbuffer;
1519 parent->t_c = ktime_get_ns(); 1537 parent->t_c = ktime_get_ns();
1520 parent->cmode = HTB_CAN_SEND; 1538 parent->cmode = HTB_CAN_SEND;
1521 if (q->offload) 1539 if (q->offload)
1522 parent->leaf.offload_queue = 1540 parent->leaf.offload_queue = cl->leaf.offload_queue;
1523 } 1541 }
1524 1542
1525 static void htb_parent_to_leaf_offload(struct 1543 static void htb_parent_to_leaf_offload(struct Qdisc *sch,
1526 struct 1544 struct netdev_queue *dev_queue,
1527 struct 1545 struct Qdisc *new_q)
1528 { 1546 {
1529 struct Qdisc *old_q; 1547 struct Qdisc *old_q;
1530 1548
1531 /* One ref for cl->leaf.q, the other 1549 /* One ref for cl->leaf.q, the other for dev_queue->qdisc. */
1532 if (new_q) 1550 if (new_q)
1533 qdisc_refcount_inc(new_q); 1551 qdisc_refcount_inc(new_q);
1534 old_q = htb_graft_helper(dev_queue, n 1552 old_q = htb_graft_helper(dev_queue, new_q);
1535 WARN_ON(!(old_q->flags & TCQ_F_BUILTI 1553 WARN_ON(!(old_q->flags & TCQ_F_BUILTIN));
1536 } 1554 }
1537 1555
1538 static int htb_destroy_class_offload(struct Q 1556 static int htb_destroy_class_offload(struct Qdisc *sch, struct htb_class *cl,
1539 bool las 1557 bool last_child, bool destroying,
1540 struct n 1558 struct netlink_ext_ack *extack)
1541 { 1559 {
1542 struct tc_htb_qopt_offload offload_op 1560 struct tc_htb_qopt_offload offload_opt;
1543 struct netdev_queue *dev_queue; 1561 struct netdev_queue *dev_queue;
1544 struct Qdisc *q = cl->leaf.q; 1562 struct Qdisc *q = cl->leaf.q;
1545 struct Qdisc *old; !! 1563 struct Qdisc *old = NULL;
1546 int err; 1564 int err;
1547 1565
1548 if (cl->level) 1566 if (cl->level)
1549 return -EINVAL; 1567 return -EINVAL;
1550 1568
1551 WARN_ON(!q); 1569 WARN_ON(!q);
1552 dev_queue = htb_offload_get_queue(cl) 1570 dev_queue = htb_offload_get_queue(cl);
1553 /* When destroying, caller qdisc_graf !! 1571 old = htb_graft_helper(dev_queue, NULL);
1554 * qdisc_put for the qdisc being dest !! 1572 if (destroying)
1555 * does not need to graft or qdisc_pu !! 1573 /* Before HTB is destroyed, the kernel grafts noop_qdisc to
1556 */ !! 1574 * all queues.
1557 if (!destroying) { <<
1558 old = htb_graft_helper(dev_qu <<
1559 /* Last qdisc grafted should <<
1560 * calling htb_delete. <<
1561 */ 1575 */
>> 1576 WARN_ON(!(old->flags & TCQ_F_BUILTIN));
>> 1577 else
1562 WARN_ON(old != q); 1578 WARN_ON(old != q);
1563 } <<
1564 1579
1565 if (cl->parent) { 1580 if (cl->parent) {
1566 _bstats_update(&cl->parent->b !! 1581 cl->parent->bstats_bias.bytes += q->bstats.bytes;
1567 u64_stats_read !! 1582 cl->parent->bstats_bias.packets += q->bstats.packets;
1568 u64_stats_read <<
1569 } 1583 }
1570 1584
1571 offload_opt = (struct tc_htb_qopt_off 1585 offload_opt = (struct tc_htb_qopt_offload) {
1572 .command = !last_child ? TC_H 1586 .command = !last_child ? TC_HTB_LEAF_DEL :
1573 destroying ? TC_HT 1587 destroying ? TC_HTB_LEAF_DEL_LAST_FORCE :
1574 TC_HTB_LEAF_DEL_LA 1588 TC_HTB_LEAF_DEL_LAST,
1575 .classid = cl->common.classid 1589 .classid = cl->common.classid,
1576 .extack = extack, 1590 .extack = extack,
1577 }; 1591 };
1578 err = htb_offload(qdisc_dev(sch), &of 1592 err = htb_offload(qdisc_dev(sch), &offload_opt);
1579 1593
1580 if (!destroying) { !! 1594 if (!err || destroying)
1581 if (!err) !! 1595 qdisc_put(old);
1582 qdisc_put(old); !! 1596 else
1583 else !! 1597 htb_graft_helper(dev_queue, old);
1584 htb_graft_helper(dev_ <<
1585 } <<
1586 1598
1587 if (last_child) 1599 if (last_child)
1588 return err; 1600 return err;
1589 1601
1590 if (!err && offload_opt.classid != TC 1602 if (!err && offload_opt.classid != TC_H_MIN(cl->common.classid)) {
1591 u32 classid = TC_H_MAJ(sch->h 1603 u32 classid = TC_H_MAJ(sch->handle) |
1592 TC_H_MIN(offloa 1604 TC_H_MIN(offload_opt.classid);
1593 struct htb_class *moved_cl = 1605 struct htb_class *moved_cl = htb_find(classid, sch);
1594 1606
1595 htb_offload_move_qdisc(sch, m 1607 htb_offload_move_qdisc(sch, moved_cl, cl, destroying);
1596 } 1608 }
1597 1609
1598 return err; 1610 return err;
1599 } 1611 }
1600 1612
1601 static void htb_destroy_class(struct Qdisc *s 1613 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
1602 { 1614 {
1603 if (!cl->level) { 1615 if (!cl->level) {
1604 WARN_ON(!cl->leaf.q); 1616 WARN_ON(!cl->leaf.q);
1605 qdisc_put(cl->leaf.q); 1617 qdisc_put(cl->leaf.q);
1606 } 1618 }
1607 gen_kill_estimator(&cl->rate_est); 1619 gen_kill_estimator(&cl->rate_est);
1608 tcf_block_put(cl->block); 1620 tcf_block_put(cl->block);
1609 kfree(cl); 1621 kfree(cl);
1610 } 1622 }
1611 1623
1612 static void htb_destroy(struct Qdisc *sch) 1624 static void htb_destroy(struct Qdisc *sch)
1613 { 1625 {
1614 struct net_device *dev = qdisc_dev(sc 1626 struct net_device *dev = qdisc_dev(sch);
1615 struct tc_htb_qopt_offload offload_op 1627 struct tc_htb_qopt_offload offload_opt;
1616 struct htb_sched *q = qdisc_priv(sch) 1628 struct htb_sched *q = qdisc_priv(sch);
1617 struct hlist_node *next; 1629 struct hlist_node *next;
1618 bool nonempty, changed; 1630 bool nonempty, changed;
1619 struct htb_class *cl; 1631 struct htb_class *cl;
1620 unsigned int i; 1632 unsigned int i;
1621 1633
1622 cancel_work_sync(&q->work); 1634 cancel_work_sync(&q->work);
1623 qdisc_watchdog_cancel(&q->watchdog); 1635 qdisc_watchdog_cancel(&q->watchdog);
1624 /* This line used to be after htb_des 1636 /* This line used to be after htb_destroy_class call below
1625 * and surprisingly it worked in 2.4. 1637 * and surprisingly it worked in 2.4. But it must precede it
1626 * because filter need its target cla 1638 * because filter need its target class alive to be able to call
1627 * unbind_filter on it (without Oops) 1639 * unbind_filter on it (without Oops).
1628 */ 1640 */
1629 tcf_block_put(q->block); 1641 tcf_block_put(q->block);
1630 1642
1631 for (i = 0; i < q->clhash.hashsize; i 1643 for (i = 0; i < q->clhash.hashsize; i++) {
1632 hlist_for_each_entry(cl, &q-> 1644 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
1633 tcf_block_put(cl->blo 1645 tcf_block_put(cl->block);
1634 cl->block = NULL; 1646 cl->block = NULL;
1635 } 1647 }
1636 } 1648 }
1637 1649
1638 do { 1650 do {
1639 nonempty = false; 1651 nonempty = false;
1640 changed = false; 1652 changed = false;
1641 for (i = 0; i < q->clhash.has 1653 for (i = 0; i < q->clhash.hashsize; i++) {
1642 hlist_for_each_entry_ 1654 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
1643 1655 common.hnode) {
1644 bool last_chi 1656 bool last_child;
1645 1657
1646 if (!q->offlo 1658 if (!q->offload) {
1647 htb_d 1659 htb_destroy_class(sch, cl);
1648 conti 1660 continue;
1649 } 1661 }
1650 1662
1651 nonempty = tr 1663 nonempty = true;
1652 1664
1653 if (cl->level 1665 if (cl->level)
1654 conti 1666 continue;
1655 1667
1656 changed = tru 1668 changed = true;
1657 1669
1658 last_child = 1670 last_child = htb_parent_last_child(cl);
1659 htb_destroy_c 1671 htb_destroy_class_offload(sch, cl, last_child,
1660 1672 true, NULL);
1661 qdisc_class_h 1673 qdisc_class_hash_remove(&q->clhash,
1662 1674 &cl->common);
1663 if (cl->paren 1675 if (cl->parent)
1664 cl->p 1676 cl->parent->children--;
1665 if (last_chil 1677 if (last_child)
1666 htb_p 1678 htb_parent_to_leaf(sch, cl, NULL);
1667 htb_destroy_c 1679 htb_destroy_class(sch, cl);
1668 } 1680 }
1669 } 1681 }
1670 } while (changed); 1682 } while (changed);
1671 WARN_ON(nonempty); 1683 WARN_ON(nonempty);
1672 1684
1673 qdisc_class_hash_destroy(&q->clhash); 1685 qdisc_class_hash_destroy(&q->clhash);
1674 __qdisc_reset_queue(&q->direct_queue) 1686 __qdisc_reset_queue(&q->direct_queue);
1675 1687
1676 if (q->offload) { !! 1688 if (!q->offload)
1677 offload_opt = (struct tc_htb_ !! 1689 return;
1678 .command = TC_HTB_DES !! 1690
1679 }; !! 1691 offload_opt = (struct tc_htb_qopt_offload) {
1680 htb_offload(dev, &offload_opt !! 1692 .command = TC_HTB_DESTROY,
1681 } !! 1693 };
>> 1694 htb_offload(dev, &offload_opt);
1682 1695
1683 if (!q->direct_qdiscs) 1696 if (!q->direct_qdiscs)
1684 return; 1697 return;
1685 for (i = 0; i < q->num_direct_qdiscs 1698 for (i = 0; i < q->num_direct_qdiscs && q->direct_qdiscs[i]; i++)
1686 qdisc_put(q->direct_qdiscs[i] 1699 qdisc_put(q->direct_qdiscs[i]);
1687 kfree(q->direct_qdiscs); 1700 kfree(q->direct_qdiscs);
1688 } 1701 }
1689 1702
1690 static int htb_delete(struct Qdisc *sch, unsi 1703 static int htb_delete(struct Qdisc *sch, unsigned long arg,
1691 struct netlink_ext_ack 1704 struct netlink_ext_ack *extack)
1692 { 1705 {
1693 struct htb_sched *q = qdisc_priv(sch) 1706 struct htb_sched *q = qdisc_priv(sch);
1694 struct htb_class *cl = (struct htb_cl 1707 struct htb_class *cl = (struct htb_class *)arg;
1695 struct Qdisc *new_q = NULL; 1708 struct Qdisc *new_q = NULL;
1696 int last_child = 0; 1709 int last_child = 0;
1697 int err; 1710 int err;
1698 1711
1699 /* TODO: why don't allow to delete su 1712 /* TODO: why don't allow to delete subtree ? references ? does
1700 * tc subsys guarantee us that in htb 1713 * tc subsys guarantee us that in htb_destroy it holds no class
1701 * refs so that we can remove childre 1714 * refs so that we can remove children safely there ?
1702 */ 1715 */
1703 if (cl->children || qdisc_class_in_us !! 1716 if (cl->children || cl->filter_cnt)
1704 NL_SET_ERR_MSG(extack, "HTB c <<
1705 return -EBUSY; 1717 return -EBUSY;
1706 } <<
1707 1718
1708 if (!cl->level && htb_parent_last_chi 1719 if (!cl->level && htb_parent_last_child(cl))
1709 last_child = 1; 1720 last_child = 1;
1710 1721
1711 if (q->offload) { 1722 if (q->offload) {
1712 err = htb_destroy_class_offlo 1723 err = htb_destroy_class_offload(sch, cl, last_child, false,
1713 1724 extack);
1714 if (err) 1725 if (err)
1715 return err; 1726 return err;
1716 } 1727 }
1717 1728
1718 if (last_child) { 1729 if (last_child) {
1719 struct netdev_queue *dev_queu 1730 struct netdev_queue *dev_queue = sch->dev_queue;
1720 1731
1721 if (q->offload) 1732 if (q->offload)
1722 dev_queue = htb_offlo 1733 dev_queue = htb_offload_get_queue(cl);
1723 1734
1724 new_q = qdisc_create_dflt(dev 1735 new_q = qdisc_create_dflt(dev_queue, &pfifo_qdisc_ops,
1725 cl- 1736 cl->parent->common.classid,
1726 NUL 1737 NULL);
1727 if (q->offload) !! 1738 if (q->offload) {
>> 1739 if (new_q)
>> 1740 htb_set_lockdep_class_child(new_q);
1728 htb_parent_to_leaf_of 1741 htb_parent_to_leaf_offload(sch, dev_queue, new_q);
>> 1742 }
1729 } 1743 }
1730 1744
1731 sch_tree_lock(sch); 1745 sch_tree_lock(sch);
1732 1746
1733 if (!cl->level) 1747 if (!cl->level)
1734 qdisc_purge_queue(cl->leaf.q) 1748 qdisc_purge_queue(cl->leaf.q);
1735 1749
1736 /* delete from hash and active; remai 1750 /* delete from hash and active; remainder in destroy_class */
1737 qdisc_class_hash_remove(&q->clhash, & 1751 qdisc_class_hash_remove(&q->clhash, &cl->common);
1738 if (cl->parent) 1752 if (cl->parent)
1739 cl->parent->children--; 1753 cl->parent->children--;
1740 1754
1741 if (cl->prio_activity) 1755 if (cl->prio_activity)
1742 htb_deactivate(q, cl); 1756 htb_deactivate(q, cl);
1743 1757
1744 if (cl->cmode != HTB_CAN_SEND) 1758 if (cl->cmode != HTB_CAN_SEND)
1745 htb_safe_rb_erase(&cl->pq_nod 1759 htb_safe_rb_erase(&cl->pq_node,
1746 &q->hlevel[ 1760 &q->hlevel[cl->level].wait_pq);
1747 1761
1748 if (last_child) 1762 if (last_child)
1749 htb_parent_to_leaf(sch, cl, n 1763 htb_parent_to_leaf(sch, cl, new_q);
1750 1764
1751 sch_tree_unlock(sch); 1765 sch_tree_unlock(sch);
1752 1766
1753 htb_destroy_class(sch, cl); 1767 htb_destroy_class(sch, cl);
1754 return 0; 1768 return 0;
1755 } 1769 }
1756 1770
1757 static int htb_change_class(struct Qdisc *sch 1771 static int htb_change_class(struct Qdisc *sch, u32 classid,
1758 u32 parentid, str 1772 u32 parentid, struct nlattr **tca,
1759 unsigned long *ar 1773 unsigned long *arg, struct netlink_ext_ack *extack)
1760 { 1774 {
1761 int err = -EINVAL; 1775 int err = -EINVAL;
1762 struct htb_sched *q = qdisc_priv(sch) 1776 struct htb_sched *q = qdisc_priv(sch);
1763 struct htb_class *cl = (struct htb_cl 1777 struct htb_class *cl = (struct htb_class *)*arg, *parent;
1764 struct tc_htb_qopt_offload offload_op 1778 struct tc_htb_qopt_offload offload_opt;
1765 struct nlattr *opt = tca[TCA_OPTIONS] 1779 struct nlattr *opt = tca[TCA_OPTIONS];
1766 struct nlattr *tb[TCA_HTB_MAX + 1]; 1780 struct nlattr *tb[TCA_HTB_MAX + 1];
1767 struct Qdisc *parent_qdisc = NULL; 1781 struct Qdisc *parent_qdisc = NULL;
1768 struct netdev_queue *dev_queue; 1782 struct netdev_queue *dev_queue;
1769 struct tc_htb_opt *hopt; 1783 struct tc_htb_opt *hopt;
1770 u64 rate64, ceil64; 1784 u64 rate64, ceil64;
1771 int warn = 0; 1785 int warn = 0;
1772 1786
1773 /* extract all subattrs from opt attr 1787 /* extract all subattrs from opt attr */
1774 if (!opt) 1788 if (!opt)
1775 goto failure; 1789 goto failure;
1776 1790
1777 err = nla_parse_nested_deprecated(tb, 1791 err = nla_parse_nested_deprecated(tb, TCA_HTB_MAX, opt, htb_policy,
1778 ext !! 1792 NULL);
1779 if (err < 0) 1793 if (err < 0)
1780 goto failure; 1794 goto failure;
1781 1795
1782 err = -EINVAL; 1796 err = -EINVAL;
1783 if (tb[TCA_HTB_PARMS] == NULL) 1797 if (tb[TCA_HTB_PARMS] == NULL)
1784 goto failure; 1798 goto failure;
1785 1799
1786 parent = parentid == TC_H_ROOT ? NULL 1800 parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch);
1787 1801
1788 hopt = nla_data(tb[TCA_HTB_PARMS]); 1802 hopt = nla_data(tb[TCA_HTB_PARMS]);
1789 if (!hopt->rate.rate || !hopt->ceil.r 1803 if (!hopt->rate.rate || !hopt->ceil.rate)
1790 goto failure; 1804 goto failure;
1791 1805
1792 if (q->offload) { <<
1793 /* Options not supported by t <<
1794 if (hopt->rate.overhead || ho <<
1795 NL_SET_ERR_MSG(extack <<
1796 goto failure; <<
1797 } <<
1798 if (hopt->rate.mpu || hopt->c <<
1799 NL_SET_ERR_MSG(extack <<
1800 goto failure; <<
1801 } <<
1802 } <<
1803 <<
1804 /* Keeping backward compatible with r 1806 /* Keeping backward compatible with rate_table based iproute2 tc */
1805 if (hopt->rate.linklayer == TC_LINKLA 1807 if (hopt->rate.linklayer == TC_LINKLAYER_UNAWARE)
1806 qdisc_put_rtab(qdisc_get_rtab 1808 qdisc_put_rtab(qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB],
1807 1809 NULL));
1808 1810
1809 if (hopt->ceil.linklayer == TC_LINKLA 1811 if (hopt->ceil.linklayer == TC_LINKLAYER_UNAWARE)
1810 qdisc_put_rtab(qdisc_get_rtab 1812 qdisc_put_rtab(qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB],
1811 1813 NULL));
1812 1814
1813 rate64 = tb[TCA_HTB_RATE64] ? nla_get 1815 rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0;
1814 ceil64 = tb[TCA_HTB_CEIL64] ? nla_get 1816 ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0;
1815 1817
1816 if (!cl) { /* new class 1818 if (!cl) { /* new class */
1817 struct net_device *dev = qdis 1819 struct net_device *dev = qdisc_dev(sch);
1818 struct Qdisc *new_q, *old_q; 1820 struct Qdisc *new_q, *old_q;
1819 int prio; 1821 int prio;
1820 struct { 1822 struct {
1821 struct nlattr 1823 struct nlattr nla;
1822 struct gnet_estimator 1824 struct gnet_estimator opt;
1823 } est = { 1825 } est = {
1824 .nla = { 1826 .nla = {
1825 .nla_len 1827 .nla_len = nla_attr_size(sizeof(est.opt)),
1826 .nla_type 1828 .nla_type = TCA_RATE,
1827 }, 1829 },
1828 .opt = { 1830 .opt = {
1829 /* 4s interva 1831 /* 4s interval, 16s averaging constant */
1830 .interval 1832 .interval = 2,
1831 .ewma_log 1833 .ewma_log = 2,
1832 }, 1834 },
1833 }; 1835 };
1834 1836
1835 /* check for valid classid */ 1837 /* check for valid classid */
1836 if (!classid || TC_H_MAJ(clas 1838 if (!classid || TC_H_MAJ(classid ^ sch->handle) ||
1837 htb_find(classid, sch)) 1839 htb_find(classid, sch))
1838 goto failure; 1840 goto failure;
1839 1841
1840 /* check maximal depth */ 1842 /* check maximal depth */
1841 if (parent && parent->parent 1843 if (parent && parent->parent && parent->parent->level < 2) {
1842 NL_SET_ERR_MSG_MOD(ex !! 1844 pr_err("htb: tree is too deep\n");
1843 goto failure; 1845 goto failure;
1844 } 1846 }
1845 err = -ENOBUFS; 1847 err = -ENOBUFS;
1846 cl = kzalloc(sizeof(*cl), GFP 1848 cl = kzalloc(sizeof(*cl), GFP_KERNEL);
1847 if (!cl) 1849 if (!cl)
1848 goto failure; 1850 goto failure;
1849 1851
1850 gnet_stats_basic_sync_init(&c <<
1851 gnet_stats_basic_sync_init(&c <<
1852 <<
1853 err = tcf_block_get(&cl->bloc 1852 err = tcf_block_get(&cl->block, &cl->filter_list, sch, extack);
1854 if (err) { 1853 if (err) {
1855 kfree(cl); 1854 kfree(cl);
1856 goto failure; 1855 goto failure;
1857 } 1856 }
1858 if (htb_rate_est || tca[TCA_R 1857 if (htb_rate_est || tca[TCA_RATE]) {
1859 err = gen_new_estimat 1858 err = gen_new_estimator(&cl->bstats, NULL,
1860 1859 &cl->rate_est,
1861 1860 NULL,
1862 !! 1861 qdisc_root_sleeping_running(sch),
1863 1862 tca[TCA_RATE] ? : &est.nla);
1864 if (err) 1863 if (err)
1865 goto err_bloc 1864 goto err_block_put;
1866 } 1865 }
1867 1866
1868 cl->children = 0; 1867 cl->children = 0;
1869 RB_CLEAR_NODE(&cl->pq_node); 1868 RB_CLEAR_NODE(&cl->pq_node);
1870 1869
1871 for (prio = 0; prio < TC_HTB_ 1870 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++)
1872 RB_CLEAR_NODE(&cl->no 1871 RB_CLEAR_NODE(&cl->node[prio]);
1873 1872
1874 cl->common.classid = classid; 1873 cl->common.classid = classid;
1875 1874
1876 /* Make sure nothing interrup 1875 /* Make sure nothing interrupts us in between of two
1877 * ndo_setup_tc calls. 1876 * ndo_setup_tc calls.
1878 */ 1877 */
1879 ASSERT_RTNL(); 1878 ASSERT_RTNL();
1880 1879
1881 /* create leaf qdisc early be 1880 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1882 * so that can't be used insi 1881 * so that can't be used inside of sch_tree_lock
1883 * -- thanks to Karlis Peisen 1882 * -- thanks to Karlis Peisenieks
1884 */ 1883 */
1885 if (!q->offload) { 1884 if (!q->offload) {
1886 dev_queue = sch->dev_ 1885 dev_queue = sch->dev_queue;
1887 } else if (!(parent && !paren 1886 } else if (!(parent && !parent->level)) {
1888 /* Assign a dev_queue 1887 /* Assign a dev_queue to this classid. */
1889 offload_opt = (struct 1888 offload_opt = (struct tc_htb_qopt_offload) {
1890 .command = TC 1889 .command = TC_HTB_LEAF_ALLOC_QUEUE,
1891 .classid = cl 1890 .classid = cl->common.classid,
1892 .parent_class 1891 .parent_classid = parent ?
1893 TC_H_ 1892 TC_H_MIN(parent->common.classid) :
1894 TC_HT 1893 TC_HTB_CLASSID_ROOT,
1895 .rate = max_t 1894 .rate = max_t(u64, hopt->rate.rate, rate64),
1896 .ceil = max_t 1895 .ceil = max_t(u64, hopt->ceil.rate, ceil64),
1897 .prio = hopt- <<
1898 .quantum = ho <<
1899 .extack = ext 1896 .extack = extack,
1900 }; 1897 };
1901 err = htb_offload(dev 1898 err = htb_offload(dev, &offload_opt);
1902 if (err) { 1899 if (err) {
1903 NL_SET_ERR_MS !! 1900 pr_err("htb: TC_HTB_LEAF_ALLOC_QUEUE failed with err = %d\n",
1904 !! 1901 err);
1905 goto err_kill 1902 goto err_kill_estimator;
1906 } 1903 }
1907 dev_queue = netdev_ge 1904 dev_queue = netdev_get_tx_queue(dev, offload_opt.qid);
1908 } else { /* First child. */ 1905 } else { /* First child. */
1909 dev_queue = htb_offlo 1906 dev_queue = htb_offload_get_queue(parent);
1910 old_q = htb_graft_hel 1907 old_q = htb_graft_helper(dev_queue, NULL);
1911 WARN_ON(old_q != pare 1908 WARN_ON(old_q != parent->leaf.q);
1912 offload_opt = (struct 1909 offload_opt = (struct tc_htb_qopt_offload) {
1913 .command = TC 1910 .command = TC_HTB_LEAF_TO_INNER,
1914 .classid = cl 1911 .classid = cl->common.classid,
1915 .parent_class 1912 .parent_classid =
1916 TC_H_ 1913 TC_H_MIN(parent->common.classid),
1917 .rate = max_t 1914 .rate = max_t(u64, hopt->rate.rate, rate64),
1918 .ceil = max_t 1915 .ceil = max_t(u64, hopt->ceil.rate, ceil64),
1919 .prio = hopt- <<
1920 .quantum = ho <<
1921 .extack = ext 1916 .extack = extack,
1922 }; 1917 };
1923 err = htb_offload(dev 1918 err = htb_offload(dev, &offload_opt);
1924 if (err) { 1919 if (err) {
1925 NL_SET_ERR_MS !! 1920 pr_err("htb: TC_HTB_LEAF_TO_INNER failed with err = %d\n",
1926 !! 1921 err);
1927 htb_graft_hel 1922 htb_graft_helper(dev_queue, old_q);
1928 goto err_kill 1923 goto err_kill_estimator;
1929 } 1924 }
1930 _bstats_update(&paren !! 1925 parent->bstats_bias.bytes += old_q->bstats.bytes;
1931 u64_st !! 1926 parent->bstats_bias.packets += old_q->bstats.packets;
1932 u64_st <<
1933 qdisc_put(old_q); 1927 qdisc_put(old_q);
1934 } 1928 }
1935 new_q = qdisc_create_dflt(dev 1929 new_q = qdisc_create_dflt(dev_queue, &pfifo_qdisc_ops,
1936 cla 1930 classid, NULL);
1937 if (q->offload) { 1931 if (q->offload) {
1938 /* One ref for cl->le !! 1932 if (new_q) {
1939 if (new_q) !! 1933 htb_set_lockdep_class_child(new_q);
>> 1934 /* One ref for cl->leaf.q, the other for
>> 1935 * dev_queue->qdisc.
>> 1936 */
1940 qdisc_refcoun 1937 qdisc_refcount_inc(new_q);
>> 1938 }
1941 old_q = htb_graft_hel 1939 old_q = htb_graft_helper(dev_queue, new_q);
1942 /* No qdisc_put neede 1940 /* No qdisc_put needed. */
1943 WARN_ON(!(old_q->flag 1941 WARN_ON(!(old_q->flags & TCQ_F_BUILTIN));
1944 } 1942 }
1945 sch_tree_lock(sch); 1943 sch_tree_lock(sch);
1946 if (parent && !parent->level) 1944 if (parent && !parent->level) {
1947 /* turn parent into i 1945 /* turn parent into inner node */
1948 qdisc_purge_queue(par 1946 qdisc_purge_queue(parent->leaf.q);
1949 parent_qdisc = parent 1947 parent_qdisc = parent->leaf.q;
1950 if (parent->prio_acti 1948 if (parent->prio_activity)
1951 htb_deactivat 1949 htb_deactivate(q, parent);
1952 1950
1953 /* remove from evt li 1951 /* remove from evt list because of level change */
1954 if (parent->cmode != 1952 if (parent->cmode != HTB_CAN_SEND) {
1955 htb_safe_rb_e 1953 htb_safe_rb_erase(&parent->pq_node, &q->hlevel[0].wait_pq);
1956 parent->cmode 1954 parent->cmode = HTB_CAN_SEND;
1957 } 1955 }
1958 parent->level = (pare 1956 parent->level = (parent->parent ? parent->parent->level
1959 : TC 1957 : TC_HTB_MAXDEPTH) - 1;
1960 memset(&parent->inner 1958 memset(&parent->inner, 0, sizeof(parent->inner));
1961 } 1959 }
1962 1960
1963 /* leaf (we) needs elementary 1961 /* leaf (we) needs elementary qdisc */
1964 cl->leaf.q = new_q ? new_q : 1962 cl->leaf.q = new_q ? new_q : &noop_qdisc;
1965 if (q->offload) 1963 if (q->offload)
1966 cl->leaf.offload_queu 1964 cl->leaf.offload_queue = dev_queue;
1967 1965
1968 cl->parent = parent; 1966 cl->parent = parent;
1969 1967
1970 /* set class to be in HTB_CAN 1968 /* set class to be in HTB_CAN_SEND state */
1971 cl->tokens = PSCHED_TICKS2NS( 1969 cl->tokens = PSCHED_TICKS2NS(hopt->buffer);
1972 cl->ctokens = PSCHED_TICKS2NS 1970 cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer);
1973 cl->mbuffer = 60ULL * NSEC_PE 1971 cl->mbuffer = 60ULL * NSEC_PER_SEC; /* 1min */
1974 cl->t_c = ktime_get_ns(); 1972 cl->t_c = ktime_get_ns();
1975 cl->cmode = HTB_CAN_SEND; 1973 cl->cmode = HTB_CAN_SEND;
1976 1974
1977 /* attach to the hash list an 1975 /* attach to the hash list and parent's family */
1978 qdisc_class_hash_insert(&q->c 1976 qdisc_class_hash_insert(&q->clhash, &cl->common);
1979 if (parent) 1977 if (parent)
1980 parent->children++; 1978 parent->children++;
1981 if (cl->leaf.q != &noop_qdisc 1979 if (cl->leaf.q != &noop_qdisc)
1982 qdisc_hash_add(cl->le 1980 qdisc_hash_add(cl->leaf.q, true);
1983 } else { 1981 } else {
1984 if (tca[TCA_RATE]) { 1982 if (tca[TCA_RATE]) {
1985 err = gen_replace_est 1983 err = gen_replace_estimator(&cl->bstats, NULL,
1986 1984 &cl->rate_est,
1987 1985 NULL,
1988 !! 1986 qdisc_root_sleeping_running(sch),
1989 1987 tca[TCA_RATE]);
1990 if (err) 1988 if (err)
1991 return err; 1989 return err;
1992 } 1990 }
1993 1991
1994 if (q->offload) { 1992 if (q->offload) {
1995 struct net_device *de 1993 struct net_device *dev = qdisc_dev(sch);
1996 1994
1997 offload_opt = (struct 1995 offload_opt = (struct tc_htb_qopt_offload) {
1998 .command = TC 1996 .command = TC_HTB_NODE_MODIFY,
1999 .classid = cl 1997 .classid = cl->common.classid,
2000 .rate = max_t 1998 .rate = max_t(u64, hopt->rate.rate, rate64),
2001 .ceil = max_t 1999 .ceil = max_t(u64, hopt->ceil.rate, ceil64),
2002 .prio = hopt- <<
2003 .quantum = ho <<
2004 .extack = ext 2000 .extack = extack,
2005 }; 2001 };
2006 err = htb_offload(dev 2002 err = htb_offload(dev, &offload_opt);
2007 if (err) 2003 if (err)
2008 /* Estimator 2004 /* Estimator was replaced, and rollback may fail
2009 * as well, s 2005 * as well, so we don't try to recover it, and
2010 * the estima 2006 * the estimator won't work property with the
2011 * offload an 2007 * offload anyway, because bstats are updated
2012 * only when 2008 * only when the stats are queried.
2013 */ 2009 */
2014 return err; 2010 return err;
2015 } 2011 }
2016 2012
2017 sch_tree_lock(sch); 2013 sch_tree_lock(sch);
2018 } 2014 }
2019 2015
2020 psched_ratecfg_precompute(&cl->rate, 2016 psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64);
2021 psched_ratecfg_precompute(&cl->ceil, 2017 psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64);
2022 2018
2023 /* it used to be a nasty bug here, we 2019 /* it used to be a nasty bug here, we have to check that node
2024 * is really leaf before changing cl- 2020 * is really leaf before changing cl->leaf !
2025 */ 2021 */
2026 if (!cl->level) { 2022 if (!cl->level) {
2027 u64 quantum = cl->rate.rate_b 2023 u64 quantum = cl->rate.rate_bytes_ps;
2028 2024
2029 do_div(quantum, q->rate2quant 2025 do_div(quantum, q->rate2quantum);
2030 cl->quantum = min_t(u64, quan 2026 cl->quantum = min_t(u64, quantum, INT_MAX);
2031 2027
2032 if (!hopt->quantum && cl->qua 2028 if (!hopt->quantum && cl->quantum < 1000) {
2033 warn = -1; 2029 warn = -1;
2034 cl->quantum = 1000; 2030 cl->quantum = 1000;
2035 } 2031 }
2036 if (!hopt->quantum && cl->qua 2032 if (!hopt->quantum && cl->quantum > 200000) {
2037 warn = 1; 2033 warn = 1;
2038 cl->quantum = 200000; 2034 cl->quantum = 200000;
2039 } 2035 }
2040 if (hopt->quantum) 2036 if (hopt->quantum)
2041 cl->quantum = hopt->q 2037 cl->quantum = hopt->quantum;
2042 if ((cl->prio = hopt->prio) > 2038 if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO)
2043 cl->prio = TC_HTB_NUM 2039 cl->prio = TC_HTB_NUMPRIO - 1;
2044 } 2040 }
2045 2041
2046 cl->buffer = PSCHED_TICKS2NS(hopt->bu 2042 cl->buffer = PSCHED_TICKS2NS(hopt->buffer);
2047 cl->cbuffer = PSCHED_TICKS2NS(hopt->c 2043 cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer);
2048 2044
2049 sch_tree_unlock(sch); 2045 sch_tree_unlock(sch);
2050 qdisc_put(parent_qdisc); 2046 qdisc_put(parent_qdisc);
2051 2047
2052 if (warn) 2048 if (warn)
2053 NL_SET_ERR_MSG_FMT_MOD(extack !! 2049 pr_warn("HTB: quantum of class %X is %s. Consider r2q change.\n",
2054 "quant !! 2050 cl->common.classid, (warn == -1 ? "small" : "big"));
2055 cl->co <<
2056 2051
2057 qdisc_class_hash_grow(sch, &q->clhash 2052 qdisc_class_hash_grow(sch, &q->clhash);
2058 2053
2059 *arg = (unsigned long)cl; 2054 *arg = (unsigned long)cl;
2060 return 0; 2055 return 0;
2061 2056
2062 err_kill_estimator: 2057 err_kill_estimator:
2063 gen_kill_estimator(&cl->rate_est); 2058 gen_kill_estimator(&cl->rate_est);
2064 err_block_put: 2059 err_block_put:
2065 tcf_block_put(cl->block); 2060 tcf_block_put(cl->block);
2066 kfree(cl); 2061 kfree(cl);
2067 failure: 2062 failure:
2068 return err; 2063 return err;
2069 } 2064 }
2070 2065
2071 static struct tcf_block *htb_tcf_block(struct 2066 static struct tcf_block *htb_tcf_block(struct Qdisc *sch, unsigned long arg,
2072 struct 2067 struct netlink_ext_ack *extack)
2073 { 2068 {
2074 struct htb_sched *q = qdisc_priv(sch) 2069 struct htb_sched *q = qdisc_priv(sch);
2075 struct htb_class *cl = (struct htb_cl 2070 struct htb_class *cl = (struct htb_class *)arg;
2076 2071
2077 return cl ? cl->block : q->block; 2072 return cl ? cl->block : q->block;
2078 } 2073 }
2079 2074
2080 static unsigned long htb_bind_filter(struct Q 2075 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
2081 u32 clas 2076 u32 classid)
2082 { 2077 {
2083 struct htb_class *cl = htb_find(class 2078 struct htb_class *cl = htb_find(classid, sch);
2084 2079
2085 /*if (cl && !cl->level) return 0; 2080 /*if (cl && !cl->level) return 0;
2086 * The line above used to be there to 2081 * The line above used to be there to prevent attaching filters to
2087 * leaves. But at least tc_index filt 2082 * leaves. But at least tc_index filter uses this just to get class
2088 * for other reasons so that we have 2083 * for other reasons so that we have to allow for it.
2089 * ---- 2084 * ----
2090 * 19.6.2002 As Werner explained it i 2085 * 19.6.2002 As Werner explained it is ok - bind filter is just
2091 * another way to "lock" the class - 2086 * another way to "lock" the class - unlike "get" this lock can
2092 * be broken by class during destroy 2087 * be broken by class during destroy IIUC.
2093 */ 2088 */
2094 if (cl) 2089 if (cl)
2095 qdisc_class_get(&cl->common); !! 2090 cl->filter_cnt++;
2096 return (unsigned long)cl; 2091 return (unsigned long)cl;
2097 } 2092 }
2098 2093
2099 static void htb_unbind_filter(struct Qdisc *s 2094 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
2100 { 2095 {
2101 struct htb_class *cl = (struct htb_cl 2096 struct htb_class *cl = (struct htb_class *)arg;
2102 2097
2103 qdisc_class_put(&cl->common); !! 2098 if (cl)
>> 2099 cl->filter_cnt--;
2104 } 2100 }
2105 2101
2106 static void htb_walk(struct Qdisc *sch, struc 2102 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
2107 { 2103 {
2108 struct htb_sched *q = qdisc_priv(sch) 2104 struct htb_sched *q = qdisc_priv(sch);
2109 struct htb_class *cl; 2105 struct htb_class *cl;
2110 unsigned int i; 2106 unsigned int i;
2111 2107
2112 if (arg->stop) 2108 if (arg->stop)
2113 return; 2109 return;
2114 2110
2115 for (i = 0; i < q->clhash.hashsize; i 2111 for (i = 0; i < q->clhash.hashsize; i++) {
2116 hlist_for_each_entry(cl, &q-> 2112 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
2117 if (!tc_qdisc_stats_d !! 2113 if (arg->count < arg->skip) {
>> 2114 arg->count++;
>> 2115 continue;
>> 2116 }
>> 2117 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
>> 2118 arg->stop = 1;
2118 return; 2119 return;
>> 2120 }
>> 2121 arg->count++;
2119 } 2122 }
2120 } 2123 }
2121 } 2124 }
2122 2125
2123 static const struct Qdisc_class_ops htb_class 2126 static const struct Qdisc_class_ops htb_class_ops = {
2124 .select_queue = htb_select_qu 2127 .select_queue = htb_select_queue,
2125 .graft = htb_graft, 2128 .graft = htb_graft,
2126 .leaf = htb_leaf, 2129 .leaf = htb_leaf,
2127 .qlen_notify = htb_qlen_noti 2130 .qlen_notify = htb_qlen_notify,
2128 .find = htb_search, 2131 .find = htb_search,
2129 .change = htb_change_cl 2132 .change = htb_change_class,
2130 .delete = htb_delete, 2133 .delete = htb_delete,
2131 .walk = htb_walk, 2134 .walk = htb_walk,
2132 .tcf_block = htb_tcf_block 2135 .tcf_block = htb_tcf_block,
2133 .bind_tcf = htb_bind_filt 2136 .bind_tcf = htb_bind_filter,
2134 .unbind_tcf = htb_unbind_fi 2137 .unbind_tcf = htb_unbind_filter,
2135 .dump = htb_dump_clas 2138 .dump = htb_dump_class,
2136 .dump_stats = htb_dump_clas 2139 .dump_stats = htb_dump_class_stats,
2137 }; 2140 };
2138 2141
2139 static struct Qdisc_ops htb_qdisc_ops __read_ 2142 static struct Qdisc_ops htb_qdisc_ops __read_mostly = {
2140 .cl_ops = &htb_class_op 2143 .cl_ops = &htb_class_ops,
2141 .id = "htb", 2144 .id = "htb",
2142 .priv_size = sizeof(struct 2145 .priv_size = sizeof(struct htb_sched),
2143 .enqueue = htb_enqueue, 2146 .enqueue = htb_enqueue,
2144 .dequeue = htb_dequeue, 2147 .dequeue = htb_dequeue,
2145 .peek = qdisc_peek_de 2148 .peek = qdisc_peek_dequeued,
2146 .init = htb_init, 2149 .init = htb_init,
2147 .attach = htb_attach, 2150 .attach = htb_attach,
2148 .reset = htb_reset, 2151 .reset = htb_reset,
2149 .destroy = htb_destroy, 2152 .destroy = htb_destroy,
2150 .dump = htb_dump, 2153 .dump = htb_dump,
2151 .owner = THIS_MODULE, 2154 .owner = THIS_MODULE,
2152 }; 2155 };
2153 MODULE_ALIAS_NET_SCH("htb"); <<
2154 2156
2155 static int __init htb_module_init(void) 2157 static int __init htb_module_init(void)
2156 { 2158 {
2157 return register_qdisc(&htb_qdisc_ops) 2159 return register_qdisc(&htb_qdisc_ops);
2158 } 2160 }
2159 static void __exit htb_module_exit(void) 2161 static void __exit htb_module_exit(void)
2160 { 2162 {
2161 unregister_qdisc(&htb_qdisc_ops); 2163 unregister_qdisc(&htb_qdisc_ops);
2162 } 2164 }
2163 2165
2164 module_init(htb_module_init) 2166 module_init(htb_module_init)
2165 module_exit(htb_module_exit) 2167 module_exit(htb_module_exit)
2166 MODULE_LICENSE("GPL"); 2168 MODULE_LICENSE("GPL");
2167 MODULE_DESCRIPTION("Hierarchical Token Bucket <<
2168 2169
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.