Commit | Line | Data |
---|---|---|
87990467 | 1 | /* |
1da177e4 LT |
2 | * net/sched/sch_htb.c Hierarchical token bucket, feed tree version |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Authors: Martin Devera, <devik@cdi.cz> | |
10 | * | |
11 | * Credits (in time order) for older HTB versions: | |
12 | * Stef Coene <stef.coene@docum.org> | |
13 | * HTB support at LARTC mailing list | |
10297b99 | 14 | * Ondrej Kraus, <krauso@barr.cz> |
1da177e4 LT |
15 | * found missing INIT_QDISC(htb) |
16 | * Vladimir Smelhaus, Aamer Akhter, Bert Hubert | |
17 | * helped a lot to locate nasty class stall bug | |
18 | * Andi Kleen, Jamal Hadi, Bert Hubert | |
19 | * code review and helpful comments on shaping | |
20 | * Tomasz Wrona, <tw@eter.tym.pl> | |
21 | * created test case so that I was able to fix nasty bug | |
22 | * Wilfried Weissmann | |
23 | * spotted bug in dequeue code and helped with fix | |
24 | * Jiri Fojtasek | |
25 | * fixed requeue routine | |
26 | * and many others. thanks. | |
1da177e4 | 27 | */ |
1da177e4 | 28 | #include <linux/module.h> |
47083fc0 | 29 | #include <linux/moduleparam.h> |
1da177e4 LT |
30 | #include <linux/types.h> |
31 | #include <linux/kernel.h> | |
1da177e4 | 32 | #include <linux/string.h> |
1da177e4 | 33 | #include <linux/errno.h> |
1da177e4 LT |
34 | #include <linux/skbuff.h> |
35 | #include <linux/list.h> | |
36 | #include <linux/compiler.h> | |
0ba48053 | 37 | #include <linux/rbtree.h> |
1224736d | 38 | #include <linux/workqueue.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
dc5fc579 | 40 | #include <net/netlink.h> |
292f1c7f | 41 | #include <net/sch_generic.h> |
1da177e4 | 42 | #include <net/pkt_sched.h> |
1da177e4 LT |
43 | |
44 | /* HTB algorithm. | |
45 | Author: devik@cdi.cz | |
46 | ======================================================================== | |
47 | HTB is like TBF with multiple classes. It is also similar to CBQ because | |
10297b99 | 48 | it allows to assign priority to each class in hierarchy. |
1da177e4 LT |
49 | In fact it is another implementation of Floyd's formal sharing. |
50 | ||
51 | Levels: | |
10297b99 | 52 | Each class is assigned level. Leaf has ALWAYS level 0 and root |
1da177e4 LT |
53 | classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level |
54 | one less than their parent. | |
55 | */ | |
56 | ||
47083fc0 | 57 | static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */ |
87990467 | 58 | #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */ |
1da177e4 LT |
59 | |
60 | #if HTB_VER >> 16 != TC_HTB_PROTOVER | |
61 | #error "Mismatched sch_htb.c and pkt_sch.h" | |
62 | #endif | |
63 | ||
47083fc0 JDB |
64 | /* Module parameter and sysfs export */ |
65 | module_param (htb_hysteresis, int, 0640); | |
66 | MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate"); | |
67 | ||
64153ce0 ED |
68 | static int htb_rate_est = 0; /* htb classes have a default rate estimator */ |
69 | module_param(htb_rate_est, int, 0640); | |
70 | MODULE_PARM_DESC(htb_rate_est, "setup a default rate estimator (4sec 16sec) for htb classes"); | |
71 | ||
1da177e4 LT |
72 | /* used internaly to keep status of single class */ |
73 | enum htb_cmode { | |
87990467 SH |
74 | HTB_CANT_SEND, /* class can't send and can't borrow */ |
75 | HTB_MAY_BORROW, /* class can't send but may borrow */ | |
76 | HTB_CAN_SEND /* class can send */ | |
1da177e4 LT |
77 | }; |
78 | ||
c9364636 ED |
79 | struct htb_prio { |
80 | union { | |
81 | struct rb_root row; | |
82 | struct rb_root feed; | |
83 | }; | |
84 | struct rb_node *ptr; | |
85 | /* When class changes from state 1->2 and disconnects from | |
86 | * parent's feed then we lost ptr value and start from the | |
87 | * first child again. Here we store classid of the | |
88 | * last valid ptr (used when ptr is NULL). | |
89 | */ | |
90 | u32 last_ptr_id; | |
91 | }; | |
92 | ||
ca4ec90b ED |
93 | /* interior & leaf nodes; props specific to leaves are marked L: |
94 | * To reduce false sharing, place mostly read fields at beginning, | |
95 | * and mostly written ones at the end. | |
96 | */ | |
87990467 | 97 | struct htb_class { |
f4c1f3e0 | 98 | struct Qdisc_class_common common; |
ca4ec90b ED |
99 | struct psched_ratecfg rate; |
100 | struct psched_ratecfg ceil; | |
101 | s64 buffer, cbuffer;/* token bucket depth/rate */ | |
102 | s64 mbuffer; /* max wait time */ | |
cbd37556 | 103 | u32 prio; /* these two are used only by leaves... */ |
ca4ec90b ED |
104 | int quantum; /* but stored for parent-to-leaf return */ |
105 | ||
25d8c0d5 | 106 | struct tcf_proto __rcu *filter_list; /* class attached filters */ |
ca4ec90b ED |
107 | int filter_cnt; |
108 | int refcnt; /* usage count of this class */ | |
109 | ||
110 | int level; /* our level (see above) */ | |
111 | unsigned int children; | |
112 | struct htb_class *parent; /* parent class */ | |
113 | ||
45203a3b | 114 | struct gnet_stats_rate_est64 rate_est; |
1da177e4 | 115 | |
ca4ec90b ED |
116 | /* |
117 | * Written often fields | |
118 | */ | |
119 | struct gnet_stats_basic_packed bstats; | |
120 | struct gnet_stats_queue qstats; | |
121 | struct tc_htb_xstats xstats; /* our special stats */ | |
87990467 | 122 | |
ca4ec90b ED |
123 | /* token bucket parameters */ |
124 | s64 tokens, ctokens;/* current number of tokens */ | |
125 | s64 t_c; /* checkpoint time */ | |
c19f7a34 | 126 | |
87990467 SH |
127 | union { |
128 | struct htb_class_leaf { | |
87990467 | 129 | struct list_head drop_list; |
c9364636 ED |
130 | int deficit[TC_HTB_MAXDEPTH]; |
131 | struct Qdisc *q; | |
87990467 SH |
132 | } leaf; |
133 | struct htb_class_inner { | |
c9364636 | 134 | struct htb_prio clprio[TC_HTB_NUMPRIO]; |
87990467 SH |
135 | } inner; |
136 | } un; | |
ca4ec90b | 137 | s64 pq_key; |
87990467 | 138 | |
ca4ec90b ED |
139 | int prio_activity; /* for which prios are we active */ |
140 | enum htb_cmode cmode; /* current mode of the class */ | |
141 | struct rb_node pq_node; /* node for event queue */ | |
142 | struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ | |
1da177e4 LT |
143 | }; |
144 | ||
c9364636 ED |
145 | struct htb_level { |
146 | struct rb_root wait_pq; | |
147 | struct htb_prio hprio[TC_HTB_NUMPRIO]; | |
148 | }; | |
149 | ||
87990467 | 150 | struct htb_sched { |
f4c1f3e0 | 151 | struct Qdisc_class_hash clhash; |
c9364636 ED |
152 | int defcls; /* class where unclassified flows go to */ |
153 | int rate2quantum; /* quant = rate / rate2quantum */ | |
1da177e4 | 154 | |
c9364636 | 155 | /* filters for qdisc itself */ |
25d8c0d5 | 156 | struct tcf_proto __rcu *filter_list; |
1da177e4 | 157 | |
c9364636 ED |
158 | #define HTB_WARN_TOOMANYEVENTS 0x1 |
159 | unsigned int warned; /* only one warning */ | |
160 | int direct_qlen; | |
161 | struct work_struct work; | |
1da177e4 | 162 | |
c9364636 ED |
163 | /* non shaped skbs; let them go directly thru */ |
164 | struct sk_buff_head direct_queue; | |
165 | long direct_pkts; | |
1da177e4 | 166 | |
c9364636 | 167 | struct qdisc_watchdog watchdog; |
1da177e4 | 168 | |
c9364636 ED |
169 | s64 now; /* cached dequeue time */ |
170 | struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */ | |
1da177e4 | 171 | |
c9364636 ED |
172 | /* time of nearest event per level (row) */ |
173 | s64 near_ev_cache[TC_HTB_MAXDEPTH]; | |
87990467 | 174 | |
c9364636 | 175 | int row_mask[TC_HTB_MAXDEPTH]; |
e82181de | 176 | |
c9364636 | 177 | struct htb_level hlevel[TC_HTB_MAXDEPTH]; |
1da177e4 LT |
178 | }; |
179 | ||
1da177e4 | 180 | /* find class in global hash table using given handle */ |
87990467 | 181 | static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch) |
1da177e4 LT |
182 | { |
183 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 184 | struct Qdisc_class_common *clc; |
0cef296d | 185 | |
f4c1f3e0 PM |
186 | clc = qdisc_class_find(&q->clhash, handle); |
187 | if (clc == NULL) | |
1da177e4 | 188 | return NULL; |
f4c1f3e0 | 189 | return container_of(clc, struct htb_class, common); |
1da177e4 LT |
190 | } |
191 | ||
192 | /** | |
193 | * htb_classify - classify a packet into class | |
194 | * | |
195 | * It returns NULL if the packet should be dropped or -1 if the packet | |
196 | * should be passed directly thru. In all other cases leaf class is returned. | |
197 | * We allow direct class selection by classid in priority. The we examine | |
198 | * filters in qdisc and in inner nodes (if higher filter points to the inner | |
199 | * node). If we end up with classid MAJOR:0 we enqueue the skb into special | |
10297b99 | 200 | * internal fifo (direct). These packets then go directly thru. If we still |
25985edc | 201 | * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful |
1da177e4 LT |
202 | * then finish and return direct queue. |
203 | */ | |
cc7ec456 | 204 | #define HTB_DIRECT ((struct htb_class *)-1L) |
1da177e4 | 205 | |
87990467 SH |
206 | static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, |
207 | int *qerr) | |
1da177e4 LT |
208 | { |
209 | struct htb_sched *q = qdisc_priv(sch); | |
210 | struct htb_class *cl; | |
211 | struct tcf_result res; | |
212 | struct tcf_proto *tcf; | |
213 | int result; | |
214 | ||
215 | /* allow to select class by setting skb->priority to valid classid; | |
cc7ec456 ED |
216 | * note that nfmark can be used too by attaching filter fw with no |
217 | * rules in it | |
218 | */ | |
1da177e4 | 219 | if (skb->priority == sch->handle) |
87990467 | 220 | return HTB_DIRECT; /* X:0 (direct flow) selected */ |
cc7ec456 | 221 | cl = htb_find(skb->priority, sch); |
29824310 HM |
222 | if (cl) { |
223 | if (cl->level == 0) | |
224 | return cl; | |
225 | /* Start with inner filter chain if a non-leaf class is selected */ | |
25d8c0d5 | 226 | tcf = rcu_dereference_bh(cl->filter_list); |
29824310 | 227 | } else { |
25d8c0d5 | 228 | tcf = rcu_dereference_bh(q->filter_list); |
29824310 | 229 | } |
1da177e4 | 230 | |
c27f339a | 231 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
3b3ae880 | 232 | while (tcf && (result = tc_classify(skb, tcf, &res, false)) >= 0) { |
1da177e4 LT |
233 | #ifdef CONFIG_NET_CLS_ACT |
234 | switch (result) { | |
235 | case TC_ACT_QUEUED: | |
87990467 | 236 | case TC_ACT_STOLEN: |
378a2f09 | 237 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
1da177e4 LT |
238 | case TC_ACT_SHOT: |
239 | return NULL; | |
240 | } | |
1da177e4 | 241 | #endif |
cc7ec456 ED |
242 | cl = (void *)res.class; |
243 | if (!cl) { | |
1da177e4 | 244 | if (res.classid == sch->handle) |
87990467 | 245 | return HTB_DIRECT; /* X:0 (direct flow) */ |
cc7ec456 ED |
246 | cl = htb_find(res.classid, sch); |
247 | if (!cl) | |
87990467 | 248 | break; /* filter selected invalid classid */ |
1da177e4 LT |
249 | } |
250 | if (!cl->level) | |
87990467 | 251 | return cl; /* we hit leaf; return it */ |
1da177e4 LT |
252 | |
253 | /* we have got inner class; apply inner filter chain */ | |
25d8c0d5 | 254 | tcf = rcu_dereference_bh(cl->filter_list); |
1da177e4 LT |
255 | } |
256 | /* classification failed; try to use default class */ | |
87990467 | 257 | cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch); |
1da177e4 | 258 | if (!cl || cl->level) |
87990467 | 259 | return HTB_DIRECT; /* bad default .. this is safe bet */ |
1da177e4 LT |
260 | return cl; |
261 | } | |
262 | ||
1da177e4 LT |
263 | /** |
264 | * htb_add_to_id_tree - adds class to the round robin list | |
265 | * | |
266 | * Routine adds class to the list (actually tree) sorted by classid. | |
267 | * Make sure that class is not already on such list for given prio. | |
268 | */ | |
87990467 SH |
269 | static void htb_add_to_id_tree(struct rb_root *root, |
270 | struct htb_class *cl, int prio) | |
1da177e4 LT |
271 | { |
272 | struct rb_node **p = &root->rb_node, *parent = NULL; | |
3bf72957 | 273 | |
1da177e4 | 274 | while (*p) { |
87990467 SH |
275 | struct htb_class *c; |
276 | parent = *p; | |
1da177e4 | 277 | c = rb_entry(parent, struct htb_class, node[prio]); |
3bf72957 | 278 | |
f4c1f3e0 | 279 | if (cl->common.classid > c->common.classid) |
1da177e4 | 280 | p = &parent->rb_right; |
87990467 | 281 | else |
1da177e4 LT |
282 | p = &parent->rb_left; |
283 | } | |
284 | rb_link_node(&cl->node[prio], parent, p); | |
285 | rb_insert_color(&cl->node[prio], root); | |
286 | } | |
287 | ||
288 | /** | |
289 | * htb_add_to_wait_tree - adds class to the event queue with delay | |
290 | * | |
291 | * The class is added to priority event queue to indicate that class will | |
292 | * change its mode in cl->pq_key microseconds. Make sure that class is not | |
293 | * already in the queue. | |
294 | */ | |
87990467 | 295 | static void htb_add_to_wait_tree(struct htb_sched *q, |
56b765b7 | 296 | struct htb_class *cl, s64 delay) |
1da177e4 | 297 | { |
c9364636 | 298 | struct rb_node **p = &q->hlevel[cl->level].wait_pq.rb_node, *parent = NULL; |
3bf72957 | 299 | |
fb983d45 PM |
300 | cl->pq_key = q->now + delay; |
301 | if (cl->pq_key == q->now) | |
1da177e4 LT |
302 | cl->pq_key++; |
303 | ||
304 | /* update the nearest event cache */ | |
fb983d45 | 305 | if (q->near_ev_cache[cl->level] > cl->pq_key) |
1da177e4 | 306 | q->near_ev_cache[cl->level] = cl->pq_key; |
87990467 | 307 | |
1da177e4 | 308 | while (*p) { |
87990467 SH |
309 | struct htb_class *c; |
310 | parent = *p; | |
1da177e4 | 311 | c = rb_entry(parent, struct htb_class, pq_node); |
fb983d45 | 312 | if (cl->pq_key >= c->pq_key) |
1da177e4 | 313 | p = &parent->rb_right; |
87990467 | 314 | else |
1da177e4 LT |
315 | p = &parent->rb_left; |
316 | } | |
317 | rb_link_node(&cl->pq_node, parent, p); | |
c9364636 | 318 | rb_insert_color(&cl->pq_node, &q->hlevel[cl->level].wait_pq); |
1da177e4 LT |
319 | } |
320 | ||
321 | /** | |
322 | * htb_next_rb_node - finds next node in binary tree | |
323 | * | |
324 | * When we are past last key we return NULL. | |
325 | * Average complexity is 2 steps per call. | |
326 | */ | |
3696f625 | 327 | static inline void htb_next_rb_node(struct rb_node **n) |
1da177e4 LT |
328 | { |
329 | *n = rb_next(*n); | |
330 | } | |
331 | ||
332 | /** | |
333 | * htb_add_class_to_row - add class to its row | |
334 | * | |
335 | * The class is added to row at priorities marked in mask. | |
336 | * It does nothing if mask == 0. | |
337 | */ | |
87990467 SH |
338 | static inline void htb_add_class_to_row(struct htb_sched *q, |
339 | struct htb_class *cl, int mask) | |
1da177e4 | 340 | { |
1da177e4 LT |
341 | q->row_mask[cl->level] |= mask; |
342 | while (mask) { | |
343 | int prio = ffz(~mask); | |
344 | mask &= ~(1 << prio); | |
c9364636 | 345 | htb_add_to_id_tree(&q->hlevel[cl->level].hprio[prio].row, cl, prio); |
1da177e4 LT |
346 | } |
347 | } | |
348 | ||
3696f625 SH |
349 | /* If this triggers, it is a bug in this code, but it need not be fatal */ |
350 | static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root) | |
351 | { | |
81771b3b | 352 | if (RB_EMPTY_NODE(rb)) { |
3696f625 SH |
353 | WARN_ON(1); |
354 | } else { | |
355 | rb_erase(rb, root); | |
356 | RB_CLEAR_NODE(rb); | |
357 | } | |
358 | } | |
359 | ||
360 | ||
1da177e4 LT |
361 | /** |
362 | * htb_remove_class_from_row - removes class from its row | |
363 | * | |
364 | * The class is removed from row at priorities marked in mask. | |
365 | * It does nothing if mask == 0. | |
366 | */ | |
87990467 SH |
367 | static inline void htb_remove_class_from_row(struct htb_sched *q, |
368 | struct htb_class *cl, int mask) | |
1da177e4 LT |
369 | { |
370 | int m = 0; | |
c9364636 | 371 | struct htb_level *hlevel = &q->hlevel[cl->level]; |
3bf72957 | 372 | |
1da177e4 LT |
373 | while (mask) { |
374 | int prio = ffz(~mask); | |
c9364636 | 375 | struct htb_prio *hprio = &hlevel->hprio[prio]; |
3696f625 | 376 | |
1da177e4 | 377 | mask &= ~(1 << prio); |
c9364636 ED |
378 | if (hprio->ptr == cl->node + prio) |
379 | htb_next_rb_node(&hprio->ptr); | |
3696f625 | 380 | |
c9364636 ED |
381 | htb_safe_rb_erase(cl->node + prio, &hprio->row); |
382 | if (!hprio->row.rb_node) | |
1da177e4 LT |
383 | m |= 1 << prio; |
384 | } | |
1da177e4 LT |
385 | q->row_mask[cl->level] &= ~m; |
386 | } | |
387 | ||
388 | /** | |
389 | * htb_activate_prios - creates active classe's feed chain | |
390 | * | |
391 | * The class is connected to ancestors and/or appropriate rows | |
10297b99 | 392 | * for priorities it is participating on. cl->cmode must be new |
1da177e4 LT |
393 | * (activated) mode. It does nothing if cl->prio_activity == 0. |
394 | */ | |
87990467 | 395 | static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
396 | { |
397 | struct htb_class *p = cl->parent; | |
87990467 | 398 | long m, mask = cl->prio_activity; |
1da177e4 LT |
399 | |
400 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
401 | m = mask; |
402 | while (m) { | |
1da177e4 LT |
403 | int prio = ffz(~m); |
404 | m &= ~(1 << prio); | |
87990467 | 405 | |
c9364636 | 406 | if (p->un.inner.clprio[prio].feed.rb_node) |
1da177e4 | 407 | /* parent already has its feed in use so that |
cc7ec456 ED |
408 | * reset bit in mask as parent is already ok |
409 | */ | |
1da177e4 | 410 | mask &= ~(1 << prio); |
87990467 | 411 | |
c9364636 | 412 | htb_add_to_id_tree(&p->un.inner.clprio[prio].feed, cl, prio); |
1da177e4 | 413 | } |
1da177e4 | 414 | p->prio_activity |= mask; |
87990467 SH |
415 | cl = p; |
416 | p = cl->parent; | |
3bf72957 | 417 | |
1da177e4 LT |
418 | } |
419 | if (cl->cmode == HTB_CAN_SEND && mask) | |
87990467 | 420 | htb_add_class_to_row(q, cl, mask); |
1da177e4 LT |
421 | } |
422 | ||
423 | /** | |
424 | * htb_deactivate_prios - remove class from feed chain | |
425 | * | |
10297b99 | 426 | * cl->cmode must represent old mode (before deactivation). It does |
1da177e4 LT |
427 | * nothing if cl->prio_activity == 0. Class is removed from all feed |
428 | * chains and rows. | |
429 | */ | |
430 | static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) | |
431 | { | |
432 | struct htb_class *p = cl->parent; | |
87990467 | 433 | long m, mask = cl->prio_activity; |
1da177e4 LT |
434 | |
435 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
436 | m = mask; |
437 | mask = 0; | |
1da177e4 LT |
438 | while (m) { |
439 | int prio = ffz(~m); | |
440 | m &= ~(1 << prio); | |
87990467 | 441 | |
c9364636 | 442 | if (p->un.inner.clprio[prio].ptr == cl->node + prio) { |
1da177e4 | 443 | /* we are removing child which is pointed to from |
cc7ec456 ED |
444 | * parent feed - forget the pointer but remember |
445 | * classid | |
446 | */ | |
c9364636 ED |
447 | p->un.inner.clprio[prio].last_ptr_id = cl->common.classid; |
448 | p->un.inner.clprio[prio].ptr = NULL; | |
1da177e4 | 449 | } |
87990467 | 450 | |
c9364636 ED |
451 | htb_safe_rb_erase(cl->node + prio, |
452 | &p->un.inner.clprio[prio].feed); | |
87990467 | 453 | |
c9364636 | 454 | if (!p->un.inner.clprio[prio].feed.rb_node) |
1da177e4 LT |
455 | mask |= 1 << prio; |
456 | } | |
3bf72957 | 457 | |
1da177e4 | 458 | p->prio_activity &= ~mask; |
87990467 SH |
459 | cl = p; |
460 | p = cl->parent; | |
3bf72957 | 461 | |
1da177e4 | 462 | } |
87990467 SH |
463 | if (cl->cmode == HTB_CAN_SEND && mask) |
464 | htb_remove_class_from_row(q, cl, mask); | |
1da177e4 LT |
465 | } |
466 | ||
56b765b7 | 467 | static inline s64 htb_lowater(const struct htb_class *cl) |
18a63e86 | 468 | { |
47083fc0 JDB |
469 | if (htb_hysteresis) |
470 | return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0; | |
471 | else | |
472 | return 0; | |
18a63e86 | 473 | } |
56b765b7 | 474 | static inline s64 htb_hiwater(const struct htb_class *cl) |
18a63e86 | 475 | { |
47083fc0 JDB |
476 | if (htb_hysteresis) |
477 | return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0; | |
478 | else | |
479 | return 0; | |
18a63e86 | 480 | } |
47083fc0 | 481 | |
18a63e86 | 482 | |
1da177e4 LT |
483 | /** |
484 | * htb_class_mode - computes and returns current class mode | |
485 | * | |
486 | * It computes cl's mode at time cl->t_c+diff and returns it. If mode | |
487 | * is not HTB_CAN_SEND then cl->pq_key is updated to time difference | |
10297b99 | 488 | * from now to time when cl will change its state. |
1da177e4 | 489 | * Also it is worth to note that class mode doesn't change simply |
10297b99 | 490 | * at cl->{c,}tokens == 0 but there can rather be hysteresis of |
1da177e4 LT |
491 | * 0 .. -cl->{c,}buffer range. It is meant to limit number of |
492 | * mode transitions per time unit. The speed gain is about 1/6. | |
493 | */ | |
87990467 | 494 | static inline enum htb_cmode |
56b765b7 | 495 | htb_class_mode(struct htb_class *cl, s64 *diff) |
1da177e4 | 496 | { |
56b765b7 | 497 | s64 toks; |
1da177e4 | 498 | |
87990467 SH |
499 | if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) { |
500 | *diff = -toks; | |
501 | return HTB_CANT_SEND; | |
502 | } | |
18a63e86 | 503 | |
87990467 SH |
504 | if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl)) |
505 | return HTB_CAN_SEND; | |
1da177e4 | 506 | |
87990467 SH |
507 | *diff = -toks; |
508 | return HTB_MAY_BORROW; | |
1da177e4 LT |
509 | } |
510 | ||
511 | /** | |
512 | * htb_change_class_mode - changes classe's mode | |
513 | * | |
514 | * This should be the only way how to change classe's mode under normal | |
515 | * cirsumstances. Routine will update feed lists linkage, change mode | |
516 | * and add class to the wait event queue if appropriate. New mode should | |
517 | * be different from old one and cl->pq_key has to be valid if changing | |
518 | * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). | |
519 | */ | |
87990467 | 520 | static void |
56b765b7 | 521 | htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, s64 *diff) |
87990467 SH |
522 | { |
523 | enum htb_cmode new_mode = htb_class_mode(cl, diff); | |
1da177e4 LT |
524 | |
525 | if (new_mode == cl->cmode) | |
87990467 SH |
526 | return; |
527 | ||
528 | if (cl->prio_activity) { /* not necessary: speed optimization */ | |
529 | if (cl->cmode != HTB_CANT_SEND) | |
530 | htb_deactivate_prios(q, cl); | |
1da177e4 | 531 | cl->cmode = new_mode; |
87990467 SH |
532 | if (new_mode != HTB_CANT_SEND) |
533 | htb_activate_prios(q, cl); | |
534 | } else | |
1da177e4 LT |
535 | cl->cmode = new_mode; |
536 | } | |
537 | ||
538 | /** | |
10297b99 | 539 | * htb_activate - inserts leaf cl into appropriate active feeds |
1da177e4 LT |
540 | * |
541 | * Routine learns (new) priority of leaf and activates feed chain | |
542 | * for the prio. It can be called on already active leaf safely. | |
543 | * It also adds leaf into droplist. | |
544 | */ | |
87990467 | 545 | static inline void htb_activate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 546 | { |
547b792c | 547 | WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen); |
3bf72957 | 548 | |
1da177e4 | 549 | if (!cl->prio_activity) { |
c19f7a34 | 550 | cl->prio_activity = 1 << cl->prio; |
87990467 SH |
551 | htb_activate_prios(q, cl); |
552 | list_add_tail(&cl->un.leaf.drop_list, | |
c19f7a34 | 553 | q->drops + cl->prio); |
1da177e4 LT |
554 | } |
555 | } | |
556 | ||
557 | /** | |
10297b99 | 558 | * htb_deactivate - remove leaf cl from active feeds |
1da177e4 LT |
559 | * |
560 | * Make sure that leaf is active. In the other words it can't be called | |
561 | * with non-active leaf. It also removes class from the drop list. | |
562 | */ | |
87990467 | 563 | static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 564 | { |
547b792c | 565 | WARN_ON(!cl->prio_activity); |
3bf72957 | 566 | |
87990467 | 567 | htb_deactivate_prios(q, cl); |
1da177e4 LT |
568 | cl->prio_activity = 0; |
569 | list_del_init(&cl->un.leaf.drop_list); | |
570 | } | |
571 | ||
572 | static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch) | |
573 | { | |
f30ab418 | 574 | int uninitialized_var(ret); |
87990467 SH |
575 | struct htb_sched *q = qdisc_priv(sch); |
576 | struct htb_class *cl = htb_classify(skb, sch, &ret); | |
577 | ||
578 | if (cl == HTB_DIRECT) { | |
579 | /* enqueue to helper queue */ | |
580 | if (q->direct_queue.qlen < q->direct_qlen) { | |
581 | __skb_queue_tail(&q->direct_queue, skb); | |
582 | q->direct_pkts++; | |
583 | } else { | |
17045755 | 584 | return qdisc_drop(skb, sch); |
87990467 | 585 | } |
1da177e4 | 586 | #ifdef CONFIG_NET_CLS_ACT |
87990467 | 587 | } else if (!cl) { |
c27f339a | 588 | if (ret & __NET_XMIT_BYPASS) |
25331d6c | 589 | qdisc_qstats_drop(sch); |
87990467 SH |
590 | kfree_skb(skb); |
591 | return ret; | |
1da177e4 | 592 | #endif |
378a2f09 JP |
593 | } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q)) != NET_XMIT_SUCCESS) { |
594 | if (net_xmit_drop_count(ret)) { | |
25331d6c | 595 | qdisc_qstats_drop(sch); |
378a2f09 JP |
596 | cl->qstats.drops++; |
597 | } | |
69747650 | 598 | return ret; |
87990467 | 599 | } else { |
87990467 SH |
600 | htb_activate(q, cl); |
601 | } | |
602 | ||
603 | sch->q.qlen++; | |
87990467 | 604 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
605 | } |
606 | ||
56b765b7 | 607 | static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, s64 diff) |
59e4220a | 608 | { |
56b765b7 | 609 | s64 toks = diff + cl->tokens; |
59e4220a JP |
610 | |
611 | if (toks > cl->buffer) | |
612 | toks = cl->buffer; | |
292f1c7f | 613 | toks -= (s64) psched_l2t_ns(&cl->rate, bytes); |
59e4220a JP |
614 | if (toks <= -cl->mbuffer) |
615 | toks = 1 - cl->mbuffer; | |
616 | ||
617 | cl->tokens = toks; | |
618 | } | |
619 | ||
56b765b7 | 620 | static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, s64 diff) |
59e4220a | 621 | { |
56b765b7 | 622 | s64 toks = diff + cl->ctokens; |
59e4220a JP |
623 | |
624 | if (toks > cl->cbuffer) | |
625 | toks = cl->cbuffer; | |
292f1c7f | 626 | toks -= (s64) psched_l2t_ns(&cl->ceil, bytes); |
59e4220a JP |
627 | if (toks <= -cl->mbuffer) |
628 | toks = 1 - cl->mbuffer; | |
629 | ||
630 | cl->ctokens = toks; | |
631 | } | |
632 | ||
1da177e4 LT |
633 | /** |
634 | * htb_charge_class - charges amount "bytes" to leaf and ancestors | |
635 | * | |
636 | * Routine assumes that packet "bytes" long was dequeued from leaf cl | |
637 | * borrowing from "level". It accounts bytes to ceil leaky bucket for | |
638 | * leaf and all ancestors and to rate bucket for ancestors at levels | |
639 | * "level" and higher. It also handles possible change of mode resulting | |
640 | * from the update. Note that mode can also increase here (MAY_BORROW to | |
641 | * CAN_SEND) because we can use more precise clock that event queue here. | |
642 | * In such case we remove class from event queue first. | |
643 | */ | |
87990467 | 644 | static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, |
c9726d68 | 645 | int level, struct sk_buff *skb) |
87990467 | 646 | { |
0abf77e5 | 647 | int bytes = qdisc_pkt_len(skb); |
1da177e4 | 648 | enum htb_cmode old_mode; |
56b765b7 | 649 | s64 diff; |
1da177e4 LT |
650 | |
651 | while (cl) { | |
56b765b7 | 652 | diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); |
1da177e4 | 653 | if (cl->level >= level) { |
87990467 SH |
654 | if (cl->level == level) |
655 | cl->xstats.lends++; | |
59e4220a | 656 | htb_accnt_tokens(cl, bytes, diff); |
1da177e4 LT |
657 | } else { |
658 | cl->xstats.borrows++; | |
87990467 | 659 | cl->tokens += diff; /* we moved t_c; update tokens */ |
1da177e4 | 660 | } |
59e4220a | 661 | htb_accnt_ctokens(cl, bytes, diff); |
1da177e4 | 662 | cl->t_c = q->now; |
1da177e4 | 663 | |
87990467 SH |
664 | old_mode = cl->cmode; |
665 | diff = 0; | |
666 | htb_change_class_mode(q, cl, &diff); | |
1da177e4 LT |
667 | if (old_mode != cl->cmode) { |
668 | if (old_mode != HTB_CAN_SEND) | |
c9364636 | 669 | htb_safe_rb_erase(&cl->pq_node, &q->hlevel[cl->level].wait_pq); |
1da177e4 | 670 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 671 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 672 | } |
1da177e4 | 673 | |
bfe0d029 ED |
674 | /* update basic stats except for leaves which are already updated */ |
675 | if (cl->level) | |
676 | bstats_update(&cl->bstats, skb); | |
677 | ||
1da177e4 LT |
678 | cl = cl->parent; |
679 | } | |
680 | } | |
681 | ||
682 | /** | |
683 | * htb_do_events - make mode changes to classes at the level | |
684 | * | |
fb983d45 | 685 | * Scans event queue for pending events and applies them. Returns time of |
1224736d | 686 | * next pending event (0 for no event in pq, q->now for too many events). |
fb983d45 | 687 | * Note: Applied are events whose have cl->pq_key <= q->now. |
1da177e4 | 688 | */ |
c9364636 | 689 | static s64 htb_do_events(struct htb_sched *q, const int level, |
5343a7f8 | 690 | unsigned long start) |
1da177e4 | 691 | { |
8f3ea33a | 692 | /* don't run for longer than 2 jiffies; 2 is used instead of |
cc7ec456 ED |
693 | * 1 to simplify things when jiffy is going to be incremented |
694 | * too soon | |
695 | */ | |
a73be040 | 696 | unsigned long stop_at = start + 2; |
c9364636 ED |
697 | struct rb_root *wait_pq = &q->hlevel[level].wait_pq; |
698 | ||
8f3ea33a | 699 | while (time_before(jiffies, stop_at)) { |
1da177e4 | 700 | struct htb_class *cl; |
56b765b7 | 701 | s64 diff; |
c9364636 | 702 | struct rb_node *p = rb_first(wait_pq); |
30bdbe39 | 703 | |
87990467 SH |
704 | if (!p) |
705 | return 0; | |
1da177e4 LT |
706 | |
707 | cl = rb_entry(p, struct htb_class, pq_node); | |
fb983d45 PM |
708 | if (cl->pq_key > q->now) |
709 | return cl->pq_key; | |
710 | ||
c9364636 | 711 | htb_safe_rb_erase(p, wait_pq); |
56b765b7 | 712 | diff = min_t(s64, q->now - cl->t_c, cl->mbuffer); |
87990467 | 713 | htb_change_class_mode(q, cl, &diff); |
1da177e4 | 714 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 715 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 716 | } |
1224736d JP |
717 | |
718 | /* too much load - let's continue after a break for scheduling */ | |
e82181de | 719 | if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) { |
c17988a9 | 720 | pr_warn("htb: too many events!\n"); |
e82181de JP |
721 | q->warned |= HTB_WARN_TOOMANYEVENTS; |
722 | } | |
1224736d JP |
723 | |
724 | return q->now; | |
1da177e4 LT |
725 | } |
726 | ||
727 | /* Returns class->node+prio from id-tree where classe's id is >= id. NULL | |
cc7ec456 ED |
728 | * is no such one exists. |
729 | */ | |
87990467 SH |
730 | static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n, |
731 | u32 id) | |
1da177e4 LT |
732 | { |
733 | struct rb_node *r = NULL; | |
734 | while (n) { | |
87990467 SH |
735 | struct htb_class *cl = |
736 | rb_entry(n, struct htb_class, node[prio]); | |
87990467 | 737 | |
f4c1f3e0 | 738 | if (id > cl->common.classid) { |
1da177e4 | 739 | n = n->rb_right; |
1b5c0077 | 740 | } else if (id < cl->common.classid) { |
1da177e4 LT |
741 | r = n; |
742 | n = n->rb_left; | |
1b5c0077 JP |
743 | } else { |
744 | return n; | |
1da177e4 LT |
745 | } |
746 | } | |
747 | return r; | |
748 | } | |
749 | ||
750 | /** | |
751 | * htb_lookup_leaf - returns next leaf class in DRR order | |
752 | * | |
753 | * Find leaf where current feed pointers points to. | |
754 | */ | |
c9364636 | 755 | static struct htb_class *htb_lookup_leaf(struct htb_prio *hprio, const int prio) |
1da177e4 LT |
756 | { |
757 | int i; | |
758 | struct { | |
759 | struct rb_node *root; | |
760 | struct rb_node **pptr; | |
761 | u32 *pid; | |
87990467 SH |
762 | } stk[TC_HTB_MAXDEPTH], *sp = stk; |
763 | ||
c9364636 ED |
764 | BUG_ON(!hprio->row.rb_node); |
765 | sp->root = hprio->row.rb_node; | |
766 | sp->pptr = &hprio->ptr; | |
767 | sp->pid = &hprio->last_ptr_id; | |
1da177e4 LT |
768 | |
769 | for (i = 0; i < 65535; i++) { | |
87990467 | 770 | if (!*sp->pptr && *sp->pid) { |
10297b99 | 771 | /* ptr was invalidated but id is valid - try to recover |
cc7ec456 ED |
772 | * the original or next ptr |
773 | */ | |
87990467 SH |
774 | *sp->pptr = |
775 | htb_id_find_next_upper(prio, sp->root, *sp->pid); | |
1da177e4 | 776 | } |
87990467 | 777 | *sp->pid = 0; /* ptr is valid now so that remove this hint as it |
cc7ec456 ED |
778 | * can become out of date quickly |
779 | */ | |
87990467 | 780 | if (!*sp->pptr) { /* we are at right end; rewind & go up */ |
1da177e4 | 781 | *sp->pptr = sp->root; |
87990467 | 782 | while ((*sp->pptr)->rb_left) |
1da177e4 LT |
783 | *sp->pptr = (*sp->pptr)->rb_left; |
784 | if (sp > stk) { | |
785 | sp--; | |
512bb43e JP |
786 | if (!*sp->pptr) { |
787 | WARN_ON(1); | |
87990467 | 788 | return NULL; |
512bb43e | 789 | } |
87990467 | 790 | htb_next_rb_node(sp->pptr); |
1da177e4 LT |
791 | } |
792 | } else { | |
793 | struct htb_class *cl; | |
c9364636 ED |
794 | struct htb_prio *clp; |
795 | ||
87990467 SH |
796 | cl = rb_entry(*sp->pptr, struct htb_class, node[prio]); |
797 | if (!cl->level) | |
1da177e4 | 798 | return cl; |
c9364636 ED |
799 | clp = &cl->un.inner.clprio[prio]; |
800 | (++sp)->root = clp->feed.rb_node; | |
801 | sp->pptr = &clp->ptr; | |
802 | sp->pid = &clp->last_ptr_id; | |
1da177e4 LT |
803 | } |
804 | } | |
547b792c | 805 | WARN_ON(1); |
1da177e4 LT |
806 | return NULL; |
807 | } | |
808 | ||
809 | /* dequeues packet at given priority and level; call only if | |
cc7ec456 ED |
810 | * you are sure that there is active class at prio/level |
811 | */ | |
c9364636 ED |
812 | static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, const int prio, |
813 | const int level) | |
1da177e4 LT |
814 | { |
815 | struct sk_buff *skb = NULL; | |
87990467 | 816 | struct htb_class *cl, *start; |
c9364636 ED |
817 | struct htb_level *hlevel = &q->hlevel[level]; |
818 | struct htb_prio *hprio = &hlevel->hprio[prio]; | |
819 | ||
1da177e4 | 820 | /* look initial class up in the row */ |
c9364636 | 821 | start = cl = htb_lookup_leaf(hprio, prio); |
87990467 | 822 | |
1da177e4 LT |
823 | do { |
824 | next: | |
512bb43e | 825 | if (unlikely(!cl)) |
87990467 | 826 | return NULL; |
1da177e4 LT |
827 | |
828 | /* class can be empty - it is unlikely but can be true if leaf | |
cc7ec456 ED |
829 | * qdisc drops packets in enqueue routine or if someone used |
830 | * graft operation on the leaf since last dequeue; | |
831 | * simply deactivate and skip such class | |
832 | */ | |
1da177e4 LT |
833 | if (unlikely(cl->un.leaf.q->q.qlen == 0)) { |
834 | struct htb_class *next; | |
87990467 | 835 | htb_deactivate(q, cl); |
1da177e4 LT |
836 | |
837 | /* row/level might become empty */ | |
838 | if ((q->row_mask[level] & (1 << prio)) == 0) | |
87990467 | 839 | return NULL; |
1da177e4 | 840 | |
c9364636 | 841 | next = htb_lookup_leaf(hprio, prio); |
87990467 SH |
842 | |
843 | if (cl == start) /* fix start if we just deleted it */ | |
1da177e4 LT |
844 | start = next; |
845 | cl = next; | |
846 | goto next; | |
847 | } | |
87990467 SH |
848 | |
849 | skb = cl->un.leaf.q->dequeue(cl->un.leaf.q); | |
850 | if (likely(skb != NULL)) | |
1da177e4 | 851 | break; |
633fe66e | 852 | |
b00355db | 853 | qdisc_warn_nonwc("htb", cl->un.leaf.q); |
c9364636 ED |
854 | htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr: |
855 | &q->hlevel[0].hprio[prio].ptr); | |
856 | cl = htb_lookup_leaf(hprio, prio); | |
1da177e4 LT |
857 | |
858 | } while (cl != start); | |
859 | ||
860 | if (likely(skb != NULL)) { | |
196d97f6 | 861 | bstats_update(&cl->bstats, skb); |
0abf77e5 JK |
862 | cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb); |
863 | if (cl->un.leaf.deficit[level] < 0) { | |
c19f7a34 | 864 | cl->un.leaf.deficit[level] += cl->quantum; |
c9364636 ED |
865 | htb_next_rb_node(level ? &cl->parent->un.inner.clprio[prio].ptr : |
866 | &q->hlevel[0].hprio[prio].ptr); | |
1da177e4 LT |
867 | } |
868 | /* this used to be after charge_class but this constelation | |
cc7ec456 ED |
869 | * gives us slightly better performance |
870 | */ | |
1da177e4 | 871 | if (!cl->un.leaf.q->q.qlen) |
87990467 | 872 | htb_deactivate(q, cl); |
c9726d68 | 873 | htb_charge_class(q, cl, level, skb); |
1da177e4 LT |
874 | } |
875 | return skb; | |
876 | } | |
877 | ||
1da177e4 LT |
878 | static struct sk_buff *htb_dequeue(struct Qdisc *sch) |
879 | { | |
9190b3b3 | 880 | struct sk_buff *skb; |
1da177e4 LT |
881 | struct htb_sched *q = qdisc_priv(sch); |
882 | int level; | |
5343a7f8 | 883 | s64 next_event; |
a73be040 | 884 | unsigned long start_at; |
1da177e4 LT |
885 | |
886 | /* try to dequeue direct packets as high prio (!) to minimize cpu work */ | |
87990467 SH |
887 | skb = __skb_dequeue(&q->direct_queue); |
888 | if (skb != NULL) { | |
9190b3b3 ED |
889 | ok: |
890 | qdisc_bstats_update(sch, skb); | |
fd245a4a | 891 | qdisc_unthrottled(sch); |
1da177e4 LT |
892 | sch->q.qlen--; |
893 | return skb; | |
894 | } | |
895 | ||
87990467 SH |
896 | if (!sch->q.qlen) |
897 | goto fin; | |
d2de875c | 898 | q->now = ktime_get_ns(); |
a73be040 | 899 | start_at = jiffies; |
1da177e4 | 900 | |
d2fe85da | 901 | next_event = q->now + 5LLU * NSEC_PER_SEC; |
633fe66e | 902 | |
1da177e4 LT |
903 | for (level = 0; level < TC_HTB_MAXDEPTH; level++) { |
904 | /* common case optimization - skip event handler quickly */ | |
905 | int m; | |
c9364636 | 906 | s64 event = q->near_ev_cache[level]; |
fb983d45 | 907 | |
c9364636 | 908 | if (q->now >= event) { |
a73be040 | 909 | event = htb_do_events(q, level, start_at); |
2e4b3b0e | 910 | if (!event) |
56b765b7 | 911 | event = q->now + NSEC_PER_SEC; |
2e4b3b0e | 912 | q->near_ev_cache[level] = event; |
c9364636 | 913 | } |
fb983d45 | 914 | |
c0851347 | 915 | if (next_event > event) |
fb983d45 | 916 | next_event = event; |
87990467 | 917 | |
1da177e4 LT |
918 | m = ~q->row_mask[level]; |
919 | while (m != (int)(-1)) { | |
87990467 | 920 | int prio = ffz(m); |
cc7ec456 | 921 | |
1da177e4 | 922 | m |= 1 << prio; |
87990467 | 923 | skb = htb_dequeue_tree(q, prio, level); |
9190b3b3 ED |
924 | if (likely(skb != NULL)) |
925 | goto ok; | |
1da177e4 LT |
926 | } |
927 | } | |
25331d6c | 928 | qdisc_qstats_overlimit(sch); |
56b765b7 V |
929 | if (likely(next_event > q->now)) { |
930 | if (!test_bit(__QDISC_STATE_DEACTIVATED, | |
931 | &qdisc_root_sleeping(q->watchdog.qdisc)->state)) { | |
932 | ktime_t time = ns_to_ktime(next_event); | |
933 | qdisc_throttled(q->watchdog.qdisc); | |
934 | hrtimer_start(&q->watchdog.timer, time, | |
4a8e320c | 935 | HRTIMER_MODE_ABS_PINNED); |
56b765b7 V |
936 | } |
937 | } else { | |
1224736d | 938 | schedule_work(&q->work); |
56b765b7 | 939 | } |
1da177e4 | 940 | fin: |
1da177e4 LT |
941 | return skb; |
942 | } | |
943 | ||
944 | /* try to drop from each class (by prio) until one succeed */ | |
87990467 | 945 | static unsigned int htb_drop(struct Qdisc *sch) |
1da177e4 LT |
946 | { |
947 | struct htb_sched *q = qdisc_priv(sch); | |
948 | int prio; | |
949 | ||
950 | for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) { | |
951 | struct list_head *p; | |
87990467 | 952 | list_for_each(p, q->drops + prio) { |
1da177e4 LT |
953 | struct htb_class *cl = list_entry(p, struct htb_class, |
954 | un.leaf.drop_list); | |
955 | unsigned int len; | |
87990467 SH |
956 | if (cl->un.leaf.q->ops->drop && |
957 | (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) { | |
1da177e4 LT |
958 | sch->q.qlen--; |
959 | if (!cl->un.leaf.q->q.qlen) | |
87990467 | 960 | htb_deactivate(q, cl); |
1da177e4 LT |
961 | return len; |
962 | } | |
963 | } | |
964 | } | |
965 | return 0; | |
966 | } | |
967 | ||
968 | /* reset all classes */ | |
969 | /* always caled under BH & queue lock */ | |
87990467 | 970 | static void htb_reset(struct Qdisc *sch) |
1da177e4 LT |
971 | { |
972 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 973 | struct htb_class *cl; |
f4c1f3e0 | 974 | unsigned int i; |
0cef296d | 975 | |
f4c1f3e0 | 976 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 977 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { |
1da177e4 | 978 | if (cl->level) |
87990467 | 979 | memset(&cl->un.inner, 0, sizeof(cl->un.inner)); |
1da177e4 | 980 | else { |
87990467 | 981 | if (cl->un.leaf.q) |
1da177e4 LT |
982 | qdisc_reset(cl->un.leaf.q); |
983 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); | |
984 | } | |
985 | cl->prio_activity = 0; | |
986 | cl->cmode = HTB_CAN_SEND; | |
1da177e4 LT |
987 | |
988 | } | |
989 | } | |
fb983d45 | 990 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
991 | __skb_queue_purge(&q->direct_queue); |
992 | sch->q.qlen = 0; | |
c9364636 | 993 | memset(q->hlevel, 0, sizeof(q->hlevel)); |
87990467 | 994 | memset(q->row_mask, 0, sizeof(q->row_mask)); |
1da177e4 | 995 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 996 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 LT |
997 | } |
998 | ||
27a3421e PM |
999 | static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = { |
1000 | [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) }, | |
1001 | [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) }, | |
1002 | [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
1003 | [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
6906f4ed | 1004 | [TCA_HTB_DIRECT_QLEN] = { .type = NLA_U32 }, |
df62cdf3 ED |
1005 | [TCA_HTB_RATE64] = { .type = NLA_U64 }, |
1006 | [TCA_HTB_CEIL64] = { .type = NLA_U64 }, | |
27a3421e PM |
1007 | }; |
1008 | ||
1224736d JP |
1009 | static void htb_work_func(struct work_struct *work) |
1010 | { | |
1011 | struct htb_sched *q = container_of(work, struct htb_sched, work); | |
1012 | struct Qdisc *sch = q->watchdog.qdisc; | |
1013 | ||
1014 | __netif_schedule(qdisc_root(sch)); | |
1015 | } | |
1016 | ||
1e90474c | 1017 | static int htb_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
1018 | { |
1019 | struct htb_sched *q = qdisc_priv(sch); | |
6906f4ed | 1020 | struct nlattr *tb[TCA_HTB_MAX + 1]; |
1da177e4 | 1021 | struct tc_htb_glob *gopt; |
cee63723 | 1022 | int err; |
1da177e4 | 1023 | int i; |
cee63723 PM |
1024 | |
1025 | if (!opt) | |
1026 | return -EINVAL; | |
1027 | ||
6906f4ed | 1028 | err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy); |
cee63723 PM |
1029 | if (err < 0) |
1030 | return err; | |
1031 | ||
6906f4ed | 1032 | if (!tb[TCA_HTB_INIT]) |
1da177e4 | 1033 | return -EINVAL; |
6906f4ed | 1034 | |
1e90474c | 1035 | gopt = nla_data(tb[TCA_HTB_INIT]); |
6906f4ed | 1036 | if (gopt->version != HTB_VER >> 16) |
1da177e4 | 1037 | return -EINVAL; |
1da177e4 | 1038 | |
f4c1f3e0 PM |
1039 | err = qdisc_class_hash_init(&q->clhash); |
1040 | if (err < 0) | |
1041 | return err; | |
1da177e4 | 1042 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 1043 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 | 1044 | |
fb983d45 | 1045 | qdisc_watchdog_init(&q->watchdog, sch); |
1224736d | 1046 | INIT_WORK(&q->work, htb_work_func); |
ab34f648 | 1047 | __skb_queue_head_init(&q->direct_queue); |
1da177e4 | 1048 | |
6906f4ed ED |
1049 | if (tb[TCA_HTB_DIRECT_QLEN]) |
1050 | q->direct_qlen = nla_get_u32(tb[TCA_HTB_DIRECT_QLEN]); | |
348e3435 | 1051 | else |
6906f4ed | 1052 | q->direct_qlen = qdisc_dev(sch)->tx_queue_len; |
348e3435 | 1053 | |
1da177e4 LT |
1054 | if ((q->rate2quantum = gopt->rate2quantum) < 1) |
1055 | q->rate2quantum = 1; | |
1056 | q->defcls = gopt->defcls; | |
1057 | ||
1058 | return 0; | |
1059 | } | |
1060 | ||
1061 | static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) | |
1062 | { | |
1063 | struct htb_sched *q = qdisc_priv(sch); | |
4b3550ef | 1064 | struct nlattr *nest; |
1da177e4 | 1065 | struct tc_htb_glob gopt; |
4b3550ef | 1066 | |
6f542efc ED |
1067 | /* Its safe to not acquire qdisc lock. As we hold RTNL, |
1068 | * no change can happen on the qdisc parameters. | |
1069 | */ | |
1da177e4 | 1070 | |
4b3550ef | 1071 | gopt.direct_pkts = q->direct_pkts; |
1da177e4 LT |
1072 | gopt.version = HTB_VER; |
1073 | gopt.rate2quantum = q->rate2quantum; | |
1074 | gopt.defcls = q->defcls; | |
3bf72957 | 1075 | gopt.debug = 0; |
4b3550ef PM |
1076 | |
1077 | nest = nla_nest_start(skb, TCA_OPTIONS); | |
1078 | if (nest == NULL) | |
1079 | goto nla_put_failure; | |
6906f4ed ED |
1080 | if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt) || |
1081 | nla_put_u32(skb, TCA_HTB_DIRECT_QLEN, q->direct_qlen)) | |
1b34ec43 | 1082 | goto nla_put_failure; |
4b3550ef | 1083 | |
6f542efc | 1084 | return nla_nest_end(skb, nest); |
4b3550ef | 1085 | |
1e90474c | 1086 | nla_put_failure: |
4b3550ef | 1087 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1088 | return -1; |
1089 | } | |
1090 | ||
1091 | static int htb_dump_class(struct Qdisc *sch, unsigned long arg, | |
87990467 | 1092 | struct sk_buff *skb, struct tcmsg *tcm) |
1da177e4 | 1093 | { |
87990467 | 1094 | struct htb_class *cl = (struct htb_class *)arg; |
4b3550ef | 1095 | struct nlattr *nest; |
1da177e4 LT |
1096 | struct tc_htb_opt opt; |
1097 | ||
6f542efc ED |
1098 | /* Its safe to not acquire qdisc lock. As we hold RTNL, |
1099 | * no change can happen on the class parameters. | |
1100 | */ | |
f4c1f3e0 PM |
1101 | tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT; |
1102 | tcm->tcm_handle = cl->common.classid; | |
1da177e4 LT |
1103 | if (!cl->level && cl->un.leaf.q) |
1104 | tcm->tcm_info = cl->un.leaf.q->handle; | |
1105 | ||
4b3550ef PM |
1106 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1107 | if (nest == NULL) | |
1108 | goto nla_put_failure; | |
1da177e4 | 1109 | |
87990467 | 1110 | memset(&opt, 0, sizeof(opt)); |
1da177e4 | 1111 | |
01cb71d2 | 1112 | psched_ratecfg_getrate(&opt.rate, &cl->rate); |
9c10f411 | 1113 | opt.buffer = PSCHED_NS2TICKS(cl->buffer); |
01cb71d2 | 1114 | psched_ratecfg_getrate(&opt.ceil, &cl->ceil); |
9c10f411 | 1115 | opt.cbuffer = PSCHED_NS2TICKS(cl->cbuffer); |
c19f7a34 JP |
1116 | opt.quantum = cl->quantum; |
1117 | opt.prio = cl->prio; | |
87990467 | 1118 | opt.level = cl->level; |
1b34ec43 DM |
1119 | if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt)) |
1120 | goto nla_put_failure; | |
df62cdf3 ED |
1121 | if ((cl->rate.rate_bytes_ps >= (1ULL << 32)) && |
1122 | nla_put_u64(skb, TCA_HTB_RATE64, cl->rate.rate_bytes_ps)) | |
1123 | goto nla_put_failure; | |
1124 | if ((cl->ceil.rate_bytes_ps >= (1ULL << 32)) && | |
1125 | nla_put_u64(skb, TCA_HTB_CEIL64, cl->ceil.rate_bytes_ps)) | |
1126 | goto nla_put_failure; | |
4b3550ef | 1127 | |
6f542efc | 1128 | return nla_nest_end(skb, nest); |
4b3550ef | 1129 | |
1e90474c | 1130 | nla_put_failure: |
4b3550ef | 1131 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1132 | return -1; |
1133 | } | |
1134 | ||
1135 | static int | |
87990467 | 1136 | htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d) |
1da177e4 | 1137 | { |
87990467 | 1138 | struct htb_class *cl = (struct htb_class *)arg; |
64015853 | 1139 | __u32 qlen = 0; |
1da177e4 | 1140 | |
1da177e4 | 1141 | if (!cl->level && cl->un.leaf.q) |
64015853 | 1142 | qlen = cl->un.leaf.q->q.qlen; |
5343a7f8 ED |
1143 | cl->xstats.tokens = PSCHED_NS2TICKS(cl->tokens); |
1144 | cl->xstats.ctokens = PSCHED_NS2TICKS(cl->ctokens); | |
1da177e4 | 1145 | |
22e0f8b9 | 1146 | if (gnet_stats_copy_basic(d, NULL, &cl->bstats) < 0 || |
d250a5f9 | 1147 | gnet_stats_copy_rate_est(d, NULL, &cl->rate_est) < 0 || |
b0ab6f92 | 1148 | gnet_stats_copy_queue(d, NULL, &cl->qstats, qlen) < 0) |
1da177e4 LT |
1149 | return -1; |
1150 | ||
1151 | return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); | |
1152 | } | |
1153 | ||
1154 | static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
87990467 | 1155 | struct Qdisc **old) |
1da177e4 | 1156 | { |
87990467 | 1157 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 | 1158 | |
5b9a9ccf PM |
1159 | if (cl->level) |
1160 | return -EINVAL; | |
1161 | if (new == NULL && | |
3511c913 | 1162 | (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
5b9a9ccf PM |
1163 | cl->common.classid)) == NULL) |
1164 | return -ENOBUFS; | |
1165 | ||
86a7996c | 1166 | *old = qdisc_replace(sch, new, &cl->un.leaf.q); |
5b9a9ccf | 1167 | return 0; |
1da177e4 LT |
1168 | } |
1169 | ||
87990467 | 1170 | static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg) |
1da177e4 | 1171 | { |
87990467 | 1172 | struct htb_class *cl = (struct htb_class *)arg; |
5b9a9ccf | 1173 | return !cl->level ? cl->un.leaf.q : NULL; |
1da177e4 LT |
1174 | } |
1175 | ||
256d61b8 PM |
1176 | static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg) |
1177 | { | |
1178 | struct htb_class *cl = (struct htb_class *)arg; | |
1179 | ||
1180 | if (cl->un.leaf.q->q.qlen == 0) | |
1181 | htb_deactivate(qdisc_priv(sch), cl); | |
1182 | } | |
1183 | ||
1da177e4 LT |
1184 | static unsigned long htb_get(struct Qdisc *sch, u32 classid) |
1185 | { | |
87990467 SH |
1186 | struct htb_class *cl = htb_find(classid, sch); |
1187 | if (cl) | |
1da177e4 LT |
1188 | cl->refcnt++; |
1189 | return (unsigned long)cl; | |
1190 | } | |
1191 | ||
160d5e10 JP |
1192 | static inline int htb_parent_last_child(struct htb_class *cl) |
1193 | { | |
1194 | if (!cl->parent) | |
1195 | /* the root class */ | |
1196 | return 0; | |
42077599 | 1197 | if (cl->parent->children > 1) |
160d5e10 JP |
1198 | /* not the last child */ |
1199 | return 0; | |
160d5e10 JP |
1200 | return 1; |
1201 | } | |
1202 | ||
3ba08b00 JP |
1203 | static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl, |
1204 | struct Qdisc *new_q) | |
160d5e10 JP |
1205 | { |
1206 | struct htb_class *parent = cl->parent; | |
1207 | ||
547b792c | 1208 | WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity); |
160d5e10 | 1209 | |
3ba08b00 | 1210 | if (parent->cmode != HTB_CAN_SEND) |
c9364636 ED |
1211 | htb_safe_rb_erase(&parent->pq_node, |
1212 | &q->hlevel[parent->level].wait_pq); | |
3ba08b00 | 1213 | |
160d5e10 JP |
1214 | parent->level = 0; |
1215 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1216 | INIT_LIST_HEAD(&parent->un.leaf.drop_list); | |
1217 | parent->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
160d5e10 JP |
1218 | parent->tokens = parent->buffer; |
1219 | parent->ctokens = parent->cbuffer; | |
d2de875c | 1220 | parent->t_c = ktime_get_ns(); |
160d5e10 JP |
1221 | parent->cmode = HTB_CAN_SEND; |
1222 | } | |
1223 | ||
87990467 | 1224 | static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl) |
1da177e4 | 1225 | { |
1da177e4 | 1226 | if (!cl->level) { |
547b792c | 1227 | WARN_ON(!cl->un.leaf.q); |
1da177e4 LT |
1228 | qdisc_destroy(cl->un.leaf.q); |
1229 | } | |
ee39e10c | 1230 | gen_kill_estimator(&cl->bstats, &cl->rate_est); |
ff31ab56 | 1231 | tcf_destroy_chain(&cl->filter_list); |
1da177e4 LT |
1232 | kfree(cl); |
1233 | } | |
1234 | ||
87990467 | 1235 | static void htb_destroy(struct Qdisc *sch) |
1da177e4 LT |
1236 | { |
1237 | struct htb_sched *q = qdisc_priv(sch); | |
b67bfe0d | 1238 | struct hlist_node *next; |
fbd8f137 PM |
1239 | struct htb_class *cl; |
1240 | unsigned int i; | |
1da177e4 | 1241 | |
1224736d | 1242 | cancel_work_sync(&q->work); |
fb983d45 | 1243 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 | 1244 | /* This line used to be after htb_destroy_class call below |
cc7ec456 ED |
1245 | * and surprisingly it worked in 2.4. But it must precede it |
1246 | * because filter need its target class alive to be able to call | |
1247 | * unbind_filter on it (without Oops). | |
1248 | */ | |
ff31ab56 | 1249 | tcf_destroy_chain(&q->filter_list); |
87990467 | 1250 | |
f4c1f3e0 | 1251 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 1252 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) |
fbd8f137 PM |
1253 | tcf_destroy_chain(&cl->filter_list); |
1254 | } | |
f4c1f3e0 | 1255 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 1256 | hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i], |
f4c1f3e0 | 1257 | common.hnode) |
fbd8f137 PM |
1258 | htb_destroy_class(sch, cl); |
1259 | } | |
f4c1f3e0 | 1260 | qdisc_class_hash_destroy(&q->clhash); |
1da177e4 LT |
1261 | __skb_queue_purge(&q->direct_queue); |
1262 | } | |
1263 | ||
1264 | static int htb_delete(struct Qdisc *sch, unsigned long arg) | |
1265 | { | |
1266 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1267 | struct htb_class *cl = (struct htb_class *)arg; |
256d61b8 | 1268 | unsigned int qlen; |
160d5e10 JP |
1269 | struct Qdisc *new_q = NULL; |
1270 | int last_child = 0; | |
1da177e4 | 1271 | |
a071d272 YY |
1272 | /* TODO: why don't allow to delete subtree ? references ? does |
1273 | * tc subsys guarantee us that in htb_destroy it holds no class | |
1274 | * refs so that we can remove children safely there ? | |
1275 | */ | |
42077599 | 1276 | if (cl->children || cl->filter_cnt) |
1da177e4 | 1277 | return -EBUSY; |
87990467 | 1278 | |
160d5e10 | 1279 | if (!cl->level && htb_parent_last_child(cl)) { |
3511c913 | 1280 | new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
bb949fbd | 1281 | cl->parent->common.classid); |
160d5e10 JP |
1282 | last_child = 1; |
1283 | } | |
1284 | ||
1da177e4 | 1285 | sch_tree_lock(sch); |
87990467 | 1286 | |
814a175e | 1287 | if (!cl->level) { |
256d61b8 | 1288 | qlen = cl->un.leaf.q->q.qlen; |
814a175e | 1289 | qdisc_reset(cl->un.leaf.q); |
256d61b8 | 1290 | qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen); |
814a175e PM |
1291 | } |
1292 | ||
f4c1f3e0 PM |
1293 | /* delete from hash and active; remainder in destroy_class */ |
1294 | qdisc_class_hash_remove(&q->clhash, &cl->common); | |
26b284de JP |
1295 | if (cl->parent) |
1296 | cl->parent->children--; | |
c38c83cb | 1297 | |
1da177e4 | 1298 | if (cl->prio_activity) |
87990467 | 1299 | htb_deactivate(q, cl); |
1da177e4 | 1300 | |
fbd8f137 | 1301 | if (cl->cmode != HTB_CAN_SEND) |
c9364636 ED |
1302 | htb_safe_rb_erase(&cl->pq_node, |
1303 | &q->hlevel[cl->level].wait_pq); | |
fbd8f137 | 1304 | |
160d5e10 | 1305 | if (last_child) |
3ba08b00 | 1306 | htb_parent_to_leaf(q, cl, new_q); |
160d5e10 | 1307 | |
7cd0a638 JP |
1308 | BUG_ON(--cl->refcnt == 0); |
1309 | /* | |
1310 | * This shouldn't happen: we "hold" one cops->get() when called | |
1311 | * from tc_ctl_tclass; the destroy method is done from cops->put(). | |
1312 | */ | |
1da177e4 LT |
1313 | |
1314 | sch_tree_unlock(sch); | |
1315 | return 0; | |
1316 | } | |
1317 | ||
1318 | static void htb_put(struct Qdisc *sch, unsigned long arg) | |
1319 | { | |
87990467 | 1320 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1321 | |
1322 | if (--cl->refcnt == 0) | |
87990467 | 1323 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1324 | } |
1325 | ||
87990467 | 1326 | static int htb_change_class(struct Qdisc *sch, u32 classid, |
1e90474c | 1327 | u32 parentid, struct nlattr **tca, |
87990467 | 1328 | unsigned long *arg) |
1da177e4 LT |
1329 | { |
1330 | int err = -EINVAL; | |
1331 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1332 | struct htb_class *cl = (struct htb_class *)*arg, *parent; |
1e90474c | 1333 | struct nlattr *opt = tca[TCA_OPTIONS]; |
6906f4ed | 1334 | struct nlattr *tb[TCA_HTB_MAX + 1]; |
1da177e4 | 1335 | struct tc_htb_opt *hopt; |
df62cdf3 | 1336 | u64 rate64, ceil64; |
1da177e4 LT |
1337 | |
1338 | /* extract all subattrs from opt attr */ | |
cee63723 PM |
1339 | if (!opt) |
1340 | goto failure; | |
1341 | ||
e18434c4 | 1342 | err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy); |
cee63723 PM |
1343 | if (err < 0) |
1344 | goto failure; | |
1345 | ||
1346 | err = -EINVAL; | |
27a3421e | 1347 | if (tb[TCA_HTB_PARMS] == NULL) |
1da177e4 | 1348 | goto failure; |
1da177e4 | 1349 | |
87990467 SH |
1350 | parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch); |
1351 | ||
1e90474c | 1352 | hopt = nla_data(tb[TCA_HTB_PARMS]); |
196d97f6 | 1353 | if (!hopt->rate.rate || !hopt->ceil.rate) |
87990467 | 1354 | goto failure; |
1da177e4 | 1355 | |
8a8e3d84 | 1356 | /* Keeping backward compatible with rate_table based iproute2 tc */ |
6b1dd856 YY |
1357 | if (hopt->rate.linklayer == TC_LINKLAYER_UNAWARE) |
1358 | qdisc_put_rtab(qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB])); | |
1359 | ||
1360 | if (hopt->ceil.linklayer == TC_LINKLAYER_UNAWARE) | |
1361 | qdisc_put_rtab(qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB])); | |
8a8e3d84 | 1362 | |
87990467 | 1363 | if (!cl) { /* new class */ |
1da177e4 | 1364 | struct Qdisc *new_q; |
3696f625 | 1365 | int prio; |
ee39e10c | 1366 | struct { |
1e90474c | 1367 | struct nlattr nla; |
ee39e10c PM |
1368 | struct gnet_estimator opt; |
1369 | } est = { | |
1e90474c PM |
1370 | .nla = { |
1371 | .nla_len = nla_attr_size(sizeof(est.opt)), | |
1372 | .nla_type = TCA_RATE, | |
ee39e10c PM |
1373 | }, |
1374 | .opt = { | |
1375 | /* 4s interval, 16s averaging constant */ | |
1376 | .interval = 2, | |
1377 | .ewma_log = 2, | |
1378 | }, | |
1379 | }; | |
3696f625 | 1380 | |
1da177e4 | 1381 | /* check for valid classid */ |
f64f9e71 JP |
1382 | if (!classid || TC_H_MAJ(classid ^ sch->handle) || |
1383 | htb_find(classid, sch)) | |
1da177e4 LT |
1384 | goto failure; |
1385 | ||
1386 | /* check maximal depth */ | |
1387 | if (parent && parent->parent && parent->parent->level < 2) { | |
cc7ec456 | 1388 | pr_err("htb: tree is too deep\n"); |
1da177e4 LT |
1389 | goto failure; |
1390 | } | |
1391 | err = -ENOBUFS; | |
cc7ec456 ED |
1392 | cl = kzalloc(sizeof(*cl), GFP_KERNEL); |
1393 | if (!cl) | |
1da177e4 | 1394 | goto failure; |
87990467 | 1395 | |
64153ce0 | 1396 | if (htb_rate_est || tca[TCA_RATE]) { |
22e0f8b9 JF |
1397 | err = gen_new_estimator(&cl->bstats, NULL, |
1398 | &cl->rate_est, | |
64153ce0 ED |
1399 | qdisc_root_sleeping_lock(sch), |
1400 | tca[TCA_RATE] ? : &est.nla); | |
1401 | if (err) { | |
1402 | kfree(cl); | |
1403 | goto failure; | |
1404 | } | |
71bcb09a SH |
1405 | } |
1406 | ||
1da177e4 | 1407 | cl->refcnt = 1; |
42077599 | 1408 | cl->children = 0; |
1da177e4 | 1409 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); |
3696f625 SH |
1410 | RB_CLEAR_NODE(&cl->pq_node); |
1411 | ||
1412 | for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) | |
1413 | RB_CLEAR_NODE(&cl->node[prio]); | |
1da177e4 LT |
1414 | |
1415 | /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL) | |
cc7ec456 ED |
1416 | * so that can't be used inside of sch_tree_lock |
1417 | * -- thanks to Karlis Peisenieks | |
1418 | */ | |
3511c913 | 1419 | new_q = qdisc_create_dflt(sch->dev_queue, |
bb949fbd | 1420 | &pfifo_qdisc_ops, classid); |
1da177e4 LT |
1421 | sch_tree_lock(sch); |
1422 | if (parent && !parent->level) { | |
256d61b8 PM |
1423 | unsigned int qlen = parent->un.leaf.q->q.qlen; |
1424 | ||
1da177e4 | 1425 | /* turn parent into inner node */ |
256d61b8 PM |
1426 | qdisc_reset(parent->un.leaf.q); |
1427 | qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen); | |
87990467 SH |
1428 | qdisc_destroy(parent->un.leaf.q); |
1429 | if (parent->prio_activity) | |
1430 | htb_deactivate(q, parent); | |
1da177e4 LT |
1431 | |
1432 | /* remove from evt list because of level change */ | |
1433 | if (parent->cmode != HTB_CAN_SEND) { | |
c9364636 | 1434 | htb_safe_rb_erase(&parent->pq_node, &q->hlevel[0].wait_pq); |
1da177e4 LT |
1435 | parent->cmode = HTB_CAN_SEND; |
1436 | } | |
1437 | parent->level = (parent->parent ? parent->parent->level | |
87990467 SH |
1438 | : TC_HTB_MAXDEPTH) - 1; |
1439 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1da177e4 LT |
1440 | } |
1441 | /* leaf (we) needs elementary qdisc */ | |
1442 | cl->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
1443 | ||
f4c1f3e0 | 1444 | cl->common.classid = classid; |
87990467 | 1445 | cl->parent = parent; |
1da177e4 LT |
1446 | |
1447 | /* set class to be in HTB_CAN_SEND state */ | |
b9a7afde JP |
1448 | cl->tokens = PSCHED_TICKS2NS(hopt->buffer); |
1449 | cl->ctokens = PSCHED_TICKS2NS(hopt->cbuffer); | |
5343a7f8 | 1450 | cl->mbuffer = 60ULL * NSEC_PER_SEC; /* 1min */ |
d2de875c | 1451 | cl->t_c = ktime_get_ns(); |
1da177e4 LT |
1452 | cl->cmode = HTB_CAN_SEND; |
1453 | ||
1454 | /* attach to the hash list and parent's family */ | |
f4c1f3e0 | 1455 | qdisc_class_hash_insert(&q->clhash, &cl->common); |
42077599 PM |
1456 | if (parent) |
1457 | parent->children++; | |
ee39e10c | 1458 | } else { |
71bcb09a | 1459 | if (tca[TCA_RATE]) { |
22e0f8b9 JF |
1460 | spinlock_t *lock = qdisc_root_sleeping_lock(sch); |
1461 | ||
1462 | err = gen_replace_estimator(&cl->bstats, NULL, | |
1463 | &cl->rate_est, | |
1464 | lock, | |
71bcb09a SH |
1465 | tca[TCA_RATE]); |
1466 | if (err) | |
1467 | return err; | |
1468 | } | |
87990467 | 1469 | sch_tree_lock(sch); |
ee39e10c | 1470 | } |
1da177e4 | 1471 | |
1598f7cb YY |
1472 | rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0; |
1473 | ||
1474 | ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0; | |
1475 | ||
1476 | psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64); | |
1477 | psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64); | |
1478 | ||
1da177e4 | 1479 | /* it used to be a nasty bug here, we have to check that node |
cc7ec456 ED |
1480 | * is really leaf before changing cl->un.leaf ! |
1481 | */ | |
1da177e4 | 1482 | if (!cl->level) { |
1598f7cb YY |
1483 | u64 quantum = cl->rate.rate_bytes_ps; |
1484 | ||
1485 | do_div(quantum, q->rate2quantum); | |
1486 | cl->quantum = min_t(u64, quantum, INT_MAX); | |
1487 | ||
c19f7a34 | 1488 | if (!hopt->quantum && cl->quantum < 1000) { |
c17988a9 YY |
1489 | pr_warn("HTB: quantum of class %X is small. Consider r2q change.\n", |
1490 | cl->common.classid); | |
c19f7a34 | 1491 | cl->quantum = 1000; |
1da177e4 | 1492 | } |
c19f7a34 | 1493 | if (!hopt->quantum && cl->quantum > 200000) { |
c17988a9 YY |
1494 | pr_warn("HTB: quantum of class %X is big. Consider r2q change.\n", |
1495 | cl->common.classid); | |
c19f7a34 | 1496 | cl->quantum = 200000; |
1da177e4 LT |
1497 | } |
1498 | if (hopt->quantum) | |
c19f7a34 JP |
1499 | cl->quantum = hopt->quantum; |
1500 | if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO) | |
1501 | cl->prio = TC_HTB_NUMPRIO - 1; | |
1da177e4 LT |
1502 | } |
1503 | ||
324f5aa5 | 1504 | cl->buffer = PSCHED_TICKS2NS(hopt->buffer); |
f3ad857e | 1505 | cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer); |
56b765b7 | 1506 | |
1da177e4 LT |
1507 | sch_tree_unlock(sch); |
1508 | ||
f4c1f3e0 PM |
1509 | qdisc_class_hash_grow(sch, &q->clhash); |
1510 | ||
1da177e4 LT |
1511 | *arg = (unsigned long)cl; |
1512 | return 0; | |
1513 | ||
1514 | failure: | |
1da177e4 LT |
1515 | return err; |
1516 | } | |
1517 | ||
25d8c0d5 JF |
1518 | static struct tcf_proto __rcu **htb_find_tcf(struct Qdisc *sch, |
1519 | unsigned long arg) | |
1da177e4 LT |
1520 | { |
1521 | struct htb_sched *q = qdisc_priv(sch); | |
1522 | struct htb_class *cl = (struct htb_class *)arg; | |
25d8c0d5 | 1523 | struct tcf_proto __rcu **fl = cl ? &cl->filter_list : &q->filter_list; |
3bf72957 | 1524 | |
1da177e4 LT |
1525 | return fl; |
1526 | } | |
1527 | ||
1528 | static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, | |
87990467 | 1529 | u32 classid) |
1da177e4 | 1530 | { |
87990467 | 1531 | struct htb_class *cl = htb_find(classid, sch); |
3bf72957 | 1532 | |
1da177e4 | 1533 | /*if (cl && !cl->level) return 0; |
cc7ec456 ED |
1534 | * The line above used to be there to prevent attaching filters to |
1535 | * leaves. But at least tc_index filter uses this just to get class | |
1536 | * for other reasons so that we have to allow for it. | |
1537 | * ---- | |
1538 | * 19.6.2002 As Werner explained it is ok - bind filter is just | |
1539 | * another way to "lock" the class - unlike "get" this lock can | |
1540 | * be broken by class during destroy IIUC. | |
1da177e4 | 1541 | */ |
87990467 SH |
1542 | if (cl) |
1543 | cl->filter_cnt++; | |
1da177e4 LT |
1544 | return (unsigned long)cl; |
1545 | } | |
1546 | ||
1547 | static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) | |
1548 | { | |
1da177e4 | 1549 | struct htb_class *cl = (struct htb_class *)arg; |
3bf72957 | 1550 | |
87990467 SH |
1551 | if (cl) |
1552 | cl->filter_cnt--; | |
1da177e4 LT |
1553 | } |
1554 | ||
1555 | static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
1556 | { | |
1557 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 1558 | struct htb_class *cl; |
f4c1f3e0 | 1559 | unsigned int i; |
1da177e4 LT |
1560 | |
1561 | if (arg->stop) | |
1562 | return; | |
1563 | ||
f4c1f3e0 | 1564 | for (i = 0; i < q->clhash.hashsize; i++) { |
b67bfe0d | 1565 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { |
1da177e4 LT |
1566 | if (arg->count < arg->skip) { |
1567 | arg->count++; | |
1568 | continue; | |
1569 | } | |
1570 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
1571 | arg->stop = 1; | |
1572 | return; | |
1573 | } | |
1574 | arg->count++; | |
1575 | } | |
1576 | } | |
1577 | } | |
1578 | ||
20fea08b | 1579 | static const struct Qdisc_class_ops htb_class_ops = { |
1da177e4 LT |
1580 | .graft = htb_graft, |
1581 | .leaf = htb_leaf, | |
256d61b8 | 1582 | .qlen_notify = htb_qlen_notify, |
1da177e4 LT |
1583 | .get = htb_get, |
1584 | .put = htb_put, | |
1585 | .change = htb_change_class, | |
1586 | .delete = htb_delete, | |
1587 | .walk = htb_walk, | |
1588 | .tcf_chain = htb_find_tcf, | |
1589 | .bind_tcf = htb_bind_filter, | |
1590 | .unbind_tcf = htb_unbind_filter, | |
1591 | .dump = htb_dump_class, | |
1592 | .dump_stats = htb_dump_class_stats, | |
1593 | }; | |
1594 | ||
20fea08b | 1595 | static struct Qdisc_ops htb_qdisc_ops __read_mostly = { |
1da177e4 LT |
1596 | .cl_ops = &htb_class_ops, |
1597 | .id = "htb", | |
1598 | .priv_size = sizeof(struct htb_sched), | |
1599 | .enqueue = htb_enqueue, | |
1600 | .dequeue = htb_dequeue, | |
77be155c | 1601 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1602 | .drop = htb_drop, |
1603 | .init = htb_init, | |
1604 | .reset = htb_reset, | |
1605 | .destroy = htb_destroy, | |
1da177e4 LT |
1606 | .dump = htb_dump, |
1607 | .owner = THIS_MODULE, | |
1608 | }; | |
1609 | ||
1610 | static int __init htb_module_init(void) | |
1611 | { | |
87990467 | 1612 | return register_qdisc(&htb_qdisc_ops); |
1da177e4 | 1613 | } |
87990467 | 1614 | static void __exit htb_module_exit(void) |
1da177e4 | 1615 | { |
87990467 | 1616 | unregister_qdisc(&htb_qdisc_ops); |
1da177e4 | 1617 | } |
87990467 | 1618 | |
1da177e4 LT |
1619 | module_init(htb_module_init) |
1620 | module_exit(htb_module_exit) | |
1621 | MODULE_LICENSE("GPL"); |