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> |
1da177e4 | 41 | #include <net/pkt_sched.h> |
1da177e4 LT |
42 | |
43 | /* HTB algorithm. | |
44 | Author: devik@cdi.cz | |
45 | ======================================================================== | |
46 | HTB is like TBF with multiple classes. It is also similar to CBQ because | |
10297b99 | 47 | it allows to assign priority to each class in hierarchy. |
1da177e4 LT |
48 | In fact it is another implementation of Floyd's formal sharing. |
49 | ||
50 | Levels: | |
10297b99 | 51 | Each class is assigned level. Leaf has ALWAYS level 0 and root |
1da177e4 LT |
52 | classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level |
53 | one less than their parent. | |
54 | */ | |
55 | ||
47083fc0 | 56 | static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */ |
87990467 | 57 | #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */ |
1da177e4 LT |
58 | |
59 | #if HTB_VER >> 16 != TC_HTB_PROTOVER | |
60 | #error "Mismatched sch_htb.c and pkt_sch.h" | |
61 | #endif | |
62 | ||
47083fc0 JDB |
63 | /* Module parameter and sysfs export */ |
64 | module_param (htb_hysteresis, int, 0640); | |
65 | MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate"); | |
66 | ||
1da177e4 LT |
67 | /* used internaly to keep status of single class */ |
68 | enum htb_cmode { | |
87990467 SH |
69 | HTB_CANT_SEND, /* class can't send and can't borrow */ |
70 | HTB_MAY_BORROW, /* class can't send but may borrow */ | |
71 | HTB_CAN_SEND /* class can send */ | |
1da177e4 LT |
72 | }; |
73 | ||
74 | /* interior & leaf nodes; props specific to leaves are marked L: */ | |
87990467 | 75 | struct htb_class { |
f4c1f3e0 | 76 | struct Qdisc_class_common common; |
87990467 | 77 | /* general class parameters */ |
c1a8f1f1 | 78 | struct gnet_stats_basic_packed bstats; |
87990467 SH |
79 | struct gnet_stats_queue qstats; |
80 | struct gnet_stats_rate_est rate_est; | |
81 | struct tc_htb_xstats xstats; /* our special stats */ | |
82 | int refcnt; /* usage count of this class */ | |
1da177e4 | 83 | |
87990467 SH |
84 | /* topology */ |
85 | int level; /* our level (see above) */ | |
42077599 | 86 | unsigned int children; |
87990467 | 87 | struct htb_class *parent; /* parent class */ |
87990467 | 88 | |
c19f7a34 JP |
89 | int prio; /* these two are used only by leaves... */ |
90 | int quantum; /* but stored for parent-to-leaf return */ | |
91 | ||
87990467 SH |
92 | union { |
93 | struct htb_class_leaf { | |
94 | struct Qdisc *q; | |
87990467 SH |
95 | int deficit[TC_HTB_MAXDEPTH]; |
96 | struct list_head drop_list; | |
97 | } leaf; | |
98 | struct htb_class_inner { | |
99 | struct rb_root feed[TC_HTB_NUMPRIO]; /* feed trees */ | |
100 | struct rb_node *ptr[TC_HTB_NUMPRIO]; /* current class ptr */ | |
101 | /* When class changes from state 1->2 and disconnects from | |
cc7ec456 ED |
102 | * parent's feed then we lost ptr value and start from the |
103 | * first child again. Here we store classid of the | |
104 | * last valid ptr (used when ptr is NULL). | |
105 | */ | |
87990467 SH |
106 | u32 last_ptr_id[TC_HTB_NUMPRIO]; |
107 | } inner; | |
108 | } un; | |
109 | struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ | |
110 | struct rb_node pq_node; /* node for event queue */ | |
fb983d45 | 111 | psched_time_t pq_key; |
87990467 SH |
112 | |
113 | int prio_activity; /* for which prios are we active */ | |
114 | enum htb_cmode cmode; /* current mode of the class */ | |
115 | ||
116 | /* class attached filters */ | |
117 | struct tcf_proto *filter_list; | |
118 | int filter_cnt; | |
119 | ||
87990467 SH |
120 | /* token bucket parameters */ |
121 | struct qdisc_rate_table *rate; /* rate table of the class itself */ | |
122 | struct qdisc_rate_table *ceil; /* ceiling rate (limits borrows too) */ | |
123 | long buffer, cbuffer; /* token bucket depth/rate */ | |
124 | psched_tdiff_t mbuffer; /* max wait time */ | |
125 | long tokens, ctokens; /* current number of tokens */ | |
126 | psched_time_t t_c; /* checkpoint time */ | |
1da177e4 LT |
127 | }; |
128 | ||
87990467 | 129 | struct htb_sched { |
f4c1f3e0 | 130 | struct Qdisc_class_hash clhash; |
0cef296d | 131 | struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */ |
87990467 SH |
132 | |
133 | /* self list - roots of self generating tree */ | |
134 | struct rb_root row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
135 | int row_mask[TC_HTB_MAXDEPTH]; | |
136 | struct rb_node *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
137 | u32 last_ptr_id[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
1da177e4 | 138 | |
87990467 SH |
139 | /* self wait list - roots of wait PQs per row */ |
140 | struct rb_root wait_pq[TC_HTB_MAXDEPTH]; | |
1da177e4 | 141 | |
87990467 | 142 | /* time of nearest event per level (row) */ |
fb983d45 | 143 | psched_time_t near_ev_cache[TC_HTB_MAXDEPTH]; |
1da177e4 | 144 | |
87990467 | 145 | int defcls; /* class where unclassified flows go to */ |
1da177e4 | 146 | |
87990467 SH |
147 | /* filters for qdisc itself */ |
148 | struct tcf_proto *filter_list; | |
1da177e4 | 149 | |
87990467 SH |
150 | int rate2quantum; /* quant = rate / rate2quantum */ |
151 | psched_time_t now; /* cached dequeue time */ | |
fb983d45 | 152 | struct qdisc_watchdog watchdog; |
1da177e4 | 153 | |
87990467 SH |
154 | /* non shaped skbs; let them go directly thru */ |
155 | struct sk_buff_head direct_queue; | |
156 | int direct_qlen; /* max qlen of above */ | |
157 | ||
158 | long direct_pkts; | |
e82181de JP |
159 | |
160 | #define HTB_WARN_TOOMANYEVENTS 0x1 | |
161 | unsigned int warned; /* only one warning */ | |
1224736d | 162 | struct work_struct work; |
1da177e4 LT |
163 | }; |
164 | ||
1da177e4 | 165 | /* find class in global hash table using given handle */ |
87990467 | 166 | static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch) |
1da177e4 LT |
167 | { |
168 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 | 169 | struct Qdisc_class_common *clc; |
0cef296d | 170 | |
f4c1f3e0 PM |
171 | clc = qdisc_class_find(&q->clhash, handle); |
172 | if (clc == NULL) | |
1da177e4 | 173 | return NULL; |
f4c1f3e0 | 174 | return container_of(clc, struct htb_class, common); |
1da177e4 LT |
175 | } |
176 | ||
177 | /** | |
178 | * htb_classify - classify a packet into class | |
179 | * | |
180 | * It returns NULL if the packet should be dropped or -1 if the packet | |
181 | * should be passed directly thru. In all other cases leaf class is returned. | |
182 | * We allow direct class selection by classid in priority. The we examine | |
183 | * filters in qdisc and in inner nodes (if higher filter points to the inner | |
184 | * node). If we end up with classid MAJOR:0 we enqueue the skb into special | |
10297b99 | 185 | * internal fifo (direct). These packets then go directly thru. If we still |
25985edc | 186 | * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful |
1da177e4 LT |
187 | * then finish and return direct queue. |
188 | */ | |
cc7ec456 | 189 | #define HTB_DIRECT ((struct htb_class *)-1L) |
1da177e4 | 190 | |
87990467 SH |
191 | static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, |
192 | int *qerr) | |
1da177e4 LT |
193 | { |
194 | struct htb_sched *q = qdisc_priv(sch); | |
195 | struct htb_class *cl; | |
196 | struct tcf_result res; | |
197 | struct tcf_proto *tcf; | |
198 | int result; | |
199 | ||
200 | /* allow to select class by setting skb->priority to valid classid; | |
cc7ec456 ED |
201 | * note that nfmark can be used too by attaching filter fw with no |
202 | * rules in it | |
203 | */ | |
1da177e4 | 204 | if (skb->priority == sch->handle) |
87990467 | 205 | return HTB_DIRECT; /* X:0 (direct flow) selected */ |
cc7ec456 ED |
206 | cl = htb_find(skb->priority, sch); |
207 | if (cl && cl->level == 0) | |
1da177e4 LT |
208 | return cl; |
209 | ||
c27f339a | 210 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 LT |
211 | tcf = q->filter_list; |
212 | while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) { | |
213 | #ifdef CONFIG_NET_CLS_ACT | |
214 | switch (result) { | |
215 | case TC_ACT_QUEUED: | |
87990467 | 216 | case TC_ACT_STOLEN: |
378a2f09 | 217 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
1da177e4 LT |
218 | case TC_ACT_SHOT: |
219 | return NULL; | |
220 | } | |
1da177e4 | 221 | #endif |
cc7ec456 ED |
222 | cl = (void *)res.class; |
223 | if (!cl) { | |
1da177e4 | 224 | if (res.classid == sch->handle) |
87990467 | 225 | return HTB_DIRECT; /* X:0 (direct flow) */ |
cc7ec456 ED |
226 | cl = htb_find(res.classid, sch); |
227 | if (!cl) | |
87990467 | 228 | break; /* filter selected invalid classid */ |
1da177e4 LT |
229 | } |
230 | if (!cl->level) | |
87990467 | 231 | return cl; /* we hit leaf; return it */ |
1da177e4 LT |
232 | |
233 | /* we have got inner class; apply inner filter chain */ | |
234 | tcf = cl->filter_list; | |
235 | } | |
236 | /* classification failed; try to use default class */ | |
87990467 | 237 | cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch); |
1da177e4 | 238 | if (!cl || cl->level) |
87990467 | 239 | return HTB_DIRECT; /* bad default .. this is safe bet */ |
1da177e4 LT |
240 | return cl; |
241 | } | |
242 | ||
1da177e4 LT |
243 | /** |
244 | * htb_add_to_id_tree - adds class to the round robin list | |
245 | * | |
246 | * Routine adds class to the list (actually tree) sorted by classid. | |
247 | * Make sure that class is not already on such list for given prio. | |
248 | */ | |
87990467 SH |
249 | static void htb_add_to_id_tree(struct rb_root *root, |
250 | struct htb_class *cl, int prio) | |
1da177e4 LT |
251 | { |
252 | struct rb_node **p = &root->rb_node, *parent = NULL; | |
3bf72957 | 253 | |
1da177e4 | 254 | while (*p) { |
87990467 SH |
255 | struct htb_class *c; |
256 | parent = *p; | |
1da177e4 | 257 | c = rb_entry(parent, struct htb_class, node[prio]); |
3bf72957 | 258 | |
f4c1f3e0 | 259 | if (cl->common.classid > c->common.classid) |
1da177e4 | 260 | p = &parent->rb_right; |
87990467 | 261 | else |
1da177e4 LT |
262 | p = &parent->rb_left; |
263 | } | |
264 | rb_link_node(&cl->node[prio], parent, p); | |
265 | rb_insert_color(&cl->node[prio], root); | |
266 | } | |
267 | ||
268 | /** | |
269 | * htb_add_to_wait_tree - adds class to the event queue with delay | |
270 | * | |
271 | * The class is added to priority event queue to indicate that class will | |
272 | * change its mode in cl->pq_key microseconds. Make sure that class is not | |
273 | * already in the queue. | |
274 | */ | |
87990467 SH |
275 | static void htb_add_to_wait_tree(struct htb_sched *q, |
276 | struct htb_class *cl, long delay) | |
1da177e4 LT |
277 | { |
278 | struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL; | |
3bf72957 | 279 | |
fb983d45 PM |
280 | cl->pq_key = q->now + delay; |
281 | if (cl->pq_key == q->now) | |
1da177e4 LT |
282 | cl->pq_key++; |
283 | ||
284 | /* update the nearest event cache */ | |
fb983d45 | 285 | if (q->near_ev_cache[cl->level] > cl->pq_key) |
1da177e4 | 286 | q->near_ev_cache[cl->level] = cl->pq_key; |
87990467 | 287 | |
1da177e4 | 288 | while (*p) { |
87990467 SH |
289 | struct htb_class *c; |
290 | parent = *p; | |
1da177e4 | 291 | c = rb_entry(parent, struct htb_class, pq_node); |
fb983d45 | 292 | if (cl->pq_key >= c->pq_key) |
1da177e4 | 293 | p = &parent->rb_right; |
87990467 | 294 | else |
1da177e4 LT |
295 | p = &parent->rb_left; |
296 | } | |
297 | rb_link_node(&cl->pq_node, parent, p); | |
298 | rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]); | |
299 | } | |
300 | ||
301 | /** | |
302 | * htb_next_rb_node - finds next node in binary tree | |
303 | * | |
304 | * When we are past last key we return NULL. | |
305 | * Average complexity is 2 steps per call. | |
306 | */ | |
3696f625 | 307 | static inline void htb_next_rb_node(struct rb_node **n) |
1da177e4 LT |
308 | { |
309 | *n = rb_next(*n); | |
310 | } | |
311 | ||
312 | /** | |
313 | * htb_add_class_to_row - add class to its row | |
314 | * | |
315 | * The class is added to row at priorities marked in mask. | |
316 | * It does nothing if mask == 0. | |
317 | */ | |
87990467 SH |
318 | static inline void htb_add_class_to_row(struct htb_sched *q, |
319 | struct htb_class *cl, int mask) | |
1da177e4 | 320 | { |
1da177e4 LT |
321 | q->row_mask[cl->level] |= mask; |
322 | while (mask) { | |
323 | int prio = ffz(~mask); | |
324 | mask &= ~(1 << prio); | |
87990467 | 325 | htb_add_to_id_tree(q->row[cl->level] + prio, cl, prio); |
1da177e4 LT |
326 | } |
327 | } | |
328 | ||
3696f625 SH |
329 | /* If this triggers, it is a bug in this code, but it need not be fatal */ |
330 | static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root) | |
331 | { | |
81771b3b | 332 | if (RB_EMPTY_NODE(rb)) { |
3696f625 SH |
333 | WARN_ON(1); |
334 | } else { | |
335 | rb_erase(rb, root); | |
336 | RB_CLEAR_NODE(rb); | |
337 | } | |
338 | } | |
339 | ||
340 | ||
1da177e4 LT |
341 | /** |
342 | * htb_remove_class_from_row - removes class from its row | |
343 | * | |
344 | * The class is removed from row at priorities marked in mask. | |
345 | * It does nothing if mask == 0. | |
346 | */ | |
87990467 SH |
347 | static inline void htb_remove_class_from_row(struct htb_sched *q, |
348 | struct htb_class *cl, int mask) | |
1da177e4 LT |
349 | { |
350 | int m = 0; | |
3bf72957 | 351 | |
1da177e4 LT |
352 | while (mask) { |
353 | int prio = ffz(~mask); | |
3696f625 | 354 | |
1da177e4 | 355 | mask &= ~(1 << prio); |
87990467 SH |
356 | if (q->ptr[cl->level][prio] == cl->node + prio) |
357 | htb_next_rb_node(q->ptr[cl->level] + prio); | |
3696f625 SH |
358 | |
359 | htb_safe_rb_erase(cl->node + prio, q->row[cl->level] + prio); | |
87990467 | 360 | if (!q->row[cl->level][prio].rb_node) |
1da177e4 LT |
361 | m |= 1 << prio; |
362 | } | |
1da177e4 LT |
363 | q->row_mask[cl->level] &= ~m; |
364 | } | |
365 | ||
366 | /** | |
367 | * htb_activate_prios - creates active classe's feed chain | |
368 | * | |
369 | * The class is connected to ancestors and/or appropriate rows | |
10297b99 | 370 | * for priorities it is participating on. cl->cmode must be new |
1da177e4 LT |
371 | * (activated) mode. It does nothing if cl->prio_activity == 0. |
372 | */ | |
87990467 | 373 | static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
374 | { |
375 | struct htb_class *p = cl->parent; | |
87990467 | 376 | long m, mask = cl->prio_activity; |
1da177e4 LT |
377 | |
378 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
379 | m = mask; |
380 | while (m) { | |
1da177e4 LT |
381 | int prio = ffz(~m); |
382 | m &= ~(1 << prio); | |
87990467 | 383 | |
1da177e4 LT |
384 | if (p->un.inner.feed[prio].rb_node) |
385 | /* parent already has its feed in use so that | |
cc7ec456 ED |
386 | * reset bit in mask as parent is already ok |
387 | */ | |
1da177e4 | 388 | mask &= ~(1 << prio); |
87990467 SH |
389 | |
390 | htb_add_to_id_tree(p->un.inner.feed + prio, cl, prio); | |
1da177e4 | 391 | } |
1da177e4 | 392 | p->prio_activity |= mask; |
87990467 SH |
393 | cl = p; |
394 | p = cl->parent; | |
3bf72957 | 395 | |
1da177e4 LT |
396 | } |
397 | if (cl->cmode == HTB_CAN_SEND && mask) | |
87990467 | 398 | htb_add_class_to_row(q, cl, mask); |
1da177e4 LT |
399 | } |
400 | ||
401 | /** | |
402 | * htb_deactivate_prios - remove class from feed chain | |
403 | * | |
10297b99 | 404 | * cl->cmode must represent old mode (before deactivation). It does |
1da177e4 LT |
405 | * nothing if cl->prio_activity == 0. Class is removed from all feed |
406 | * chains and rows. | |
407 | */ | |
408 | static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) | |
409 | { | |
410 | struct htb_class *p = cl->parent; | |
87990467 | 411 | long m, mask = cl->prio_activity; |
1da177e4 LT |
412 | |
413 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
414 | m = mask; |
415 | mask = 0; | |
1da177e4 LT |
416 | while (m) { |
417 | int prio = ffz(~m); | |
418 | m &= ~(1 << prio); | |
87990467 SH |
419 | |
420 | if (p->un.inner.ptr[prio] == cl->node + prio) { | |
1da177e4 | 421 | /* we are removing child which is pointed to from |
cc7ec456 ED |
422 | * parent feed - forget the pointer but remember |
423 | * classid | |
424 | */ | |
f4c1f3e0 | 425 | p->un.inner.last_ptr_id[prio] = cl->common.classid; |
1da177e4 LT |
426 | p->un.inner.ptr[prio] = NULL; |
427 | } | |
87990467 | 428 | |
3696f625 | 429 | htb_safe_rb_erase(cl->node + prio, p->un.inner.feed + prio); |
87990467 SH |
430 | |
431 | if (!p->un.inner.feed[prio].rb_node) | |
1da177e4 LT |
432 | mask |= 1 << prio; |
433 | } | |
3bf72957 | 434 | |
1da177e4 | 435 | p->prio_activity &= ~mask; |
87990467 SH |
436 | cl = p; |
437 | p = cl->parent; | |
3bf72957 | 438 | |
1da177e4 | 439 | } |
87990467 SH |
440 | if (cl->cmode == HTB_CAN_SEND && mask) |
441 | htb_remove_class_from_row(q, cl, mask); | |
1da177e4 LT |
442 | } |
443 | ||
18a63e86 SH |
444 | static inline long htb_lowater(const struct htb_class *cl) |
445 | { | |
47083fc0 JDB |
446 | if (htb_hysteresis) |
447 | return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0; | |
448 | else | |
449 | return 0; | |
18a63e86 SH |
450 | } |
451 | static inline long htb_hiwater(const struct htb_class *cl) | |
452 | { | |
47083fc0 JDB |
453 | if (htb_hysteresis) |
454 | return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0; | |
455 | else | |
456 | return 0; | |
18a63e86 | 457 | } |
47083fc0 | 458 | |
18a63e86 | 459 | |
1da177e4 LT |
460 | /** |
461 | * htb_class_mode - computes and returns current class mode | |
462 | * | |
463 | * It computes cl's mode at time cl->t_c+diff and returns it. If mode | |
464 | * is not HTB_CAN_SEND then cl->pq_key is updated to time difference | |
10297b99 | 465 | * from now to time when cl will change its state. |
1da177e4 | 466 | * Also it is worth to note that class mode doesn't change simply |
10297b99 | 467 | * at cl->{c,}tokens == 0 but there can rather be hysteresis of |
1da177e4 LT |
468 | * 0 .. -cl->{c,}buffer range. It is meant to limit number of |
469 | * mode transitions per time unit. The speed gain is about 1/6. | |
470 | */ | |
87990467 SH |
471 | static inline enum htb_cmode |
472 | htb_class_mode(struct htb_class *cl, long *diff) | |
1da177e4 | 473 | { |
87990467 | 474 | long toks; |
1da177e4 | 475 | |
87990467 SH |
476 | if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) { |
477 | *diff = -toks; | |
478 | return HTB_CANT_SEND; | |
479 | } | |
18a63e86 | 480 | |
87990467 SH |
481 | if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl)) |
482 | return HTB_CAN_SEND; | |
1da177e4 | 483 | |
87990467 SH |
484 | *diff = -toks; |
485 | return HTB_MAY_BORROW; | |
1da177e4 LT |
486 | } |
487 | ||
488 | /** | |
489 | * htb_change_class_mode - changes classe's mode | |
490 | * | |
491 | * This should be the only way how to change classe's mode under normal | |
492 | * cirsumstances. Routine will update feed lists linkage, change mode | |
493 | * and add class to the wait event queue if appropriate. New mode should | |
494 | * be different from old one and cl->pq_key has to be valid if changing | |
495 | * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). | |
496 | */ | |
87990467 | 497 | static void |
1da177e4 | 498 | htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff) |
87990467 SH |
499 | { |
500 | enum htb_cmode new_mode = htb_class_mode(cl, diff); | |
1da177e4 LT |
501 | |
502 | if (new_mode == cl->cmode) | |
87990467 SH |
503 | return; |
504 | ||
505 | if (cl->prio_activity) { /* not necessary: speed optimization */ | |
506 | if (cl->cmode != HTB_CANT_SEND) | |
507 | htb_deactivate_prios(q, cl); | |
1da177e4 | 508 | cl->cmode = new_mode; |
87990467 SH |
509 | if (new_mode != HTB_CANT_SEND) |
510 | htb_activate_prios(q, cl); | |
511 | } else | |
1da177e4 LT |
512 | cl->cmode = new_mode; |
513 | } | |
514 | ||
515 | /** | |
10297b99 | 516 | * htb_activate - inserts leaf cl into appropriate active feeds |
1da177e4 LT |
517 | * |
518 | * Routine learns (new) priority of leaf and activates feed chain | |
519 | * for the prio. It can be called on already active leaf safely. | |
520 | * It also adds leaf into droplist. | |
521 | */ | |
87990467 | 522 | static inline void htb_activate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 523 | { |
547b792c | 524 | WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen); |
3bf72957 | 525 | |
1da177e4 | 526 | if (!cl->prio_activity) { |
c19f7a34 | 527 | cl->prio_activity = 1 << cl->prio; |
87990467 SH |
528 | htb_activate_prios(q, cl); |
529 | list_add_tail(&cl->un.leaf.drop_list, | |
c19f7a34 | 530 | q->drops + cl->prio); |
1da177e4 LT |
531 | } |
532 | } | |
533 | ||
534 | /** | |
10297b99 | 535 | * htb_deactivate - remove leaf cl from active feeds |
1da177e4 LT |
536 | * |
537 | * Make sure that leaf is active. In the other words it can't be called | |
538 | * with non-active leaf. It also removes class from the drop list. | |
539 | */ | |
87990467 | 540 | static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 | 541 | { |
547b792c | 542 | WARN_ON(!cl->prio_activity); |
3bf72957 | 543 | |
87990467 | 544 | htb_deactivate_prios(q, cl); |
1da177e4 LT |
545 | cl->prio_activity = 0; |
546 | list_del_init(&cl->un.leaf.drop_list); | |
547 | } | |
548 | ||
549 | static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch) | |
550 | { | |
f30ab418 | 551 | int uninitialized_var(ret); |
87990467 SH |
552 | struct htb_sched *q = qdisc_priv(sch); |
553 | struct htb_class *cl = htb_classify(skb, sch, &ret); | |
554 | ||
555 | if (cl == HTB_DIRECT) { | |
556 | /* enqueue to helper queue */ | |
557 | if (q->direct_queue.qlen < q->direct_qlen) { | |
558 | __skb_queue_tail(&q->direct_queue, skb); | |
559 | q->direct_pkts++; | |
560 | } else { | |
561 | kfree_skb(skb); | |
562 | sch->qstats.drops++; | |
563 | return NET_XMIT_DROP; | |
564 | } | |
1da177e4 | 565 | #ifdef CONFIG_NET_CLS_ACT |
87990467 | 566 | } else if (!cl) { |
c27f339a | 567 | if (ret & __NET_XMIT_BYPASS) |
87990467 SH |
568 | sch->qstats.drops++; |
569 | kfree_skb(skb); | |
570 | return ret; | |
1da177e4 | 571 | #endif |
378a2f09 JP |
572 | } else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q)) != NET_XMIT_SUCCESS) { |
573 | if (net_xmit_drop_count(ret)) { | |
574 | sch->qstats.drops++; | |
575 | cl->qstats.drops++; | |
576 | } | |
69747650 | 577 | return ret; |
87990467 | 578 | } else { |
bfe0d029 | 579 | bstats_update(&cl->bstats, skb); |
87990467 SH |
580 | htb_activate(q, cl); |
581 | } | |
582 | ||
583 | sch->q.qlen++; | |
87990467 | 584 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
585 | } |
586 | ||
59e4220a JP |
587 | static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, long diff) |
588 | { | |
589 | long toks = diff + cl->tokens; | |
590 | ||
591 | if (toks > cl->buffer) | |
592 | toks = cl->buffer; | |
593 | toks -= (long) qdisc_l2t(cl->rate, bytes); | |
594 | if (toks <= -cl->mbuffer) | |
595 | toks = 1 - cl->mbuffer; | |
596 | ||
597 | cl->tokens = toks; | |
598 | } | |
599 | ||
600 | static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, long diff) | |
601 | { | |
602 | long toks = diff + cl->ctokens; | |
603 | ||
604 | if (toks > cl->cbuffer) | |
605 | toks = cl->cbuffer; | |
606 | toks -= (long) qdisc_l2t(cl->ceil, bytes); | |
607 | if (toks <= -cl->mbuffer) | |
608 | toks = 1 - cl->mbuffer; | |
609 | ||
610 | cl->ctokens = toks; | |
611 | } | |
612 | ||
1da177e4 LT |
613 | /** |
614 | * htb_charge_class - charges amount "bytes" to leaf and ancestors | |
615 | * | |
616 | * Routine assumes that packet "bytes" long was dequeued from leaf cl | |
617 | * borrowing from "level". It accounts bytes to ceil leaky bucket for | |
618 | * leaf and all ancestors and to rate bucket for ancestors at levels | |
619 | * "level" and higher. It also handles possible change of mode resulting | |
620 | * from the update. Note that mode can also increase here (MAY_BORROW to | |
621 | * CAN_SEND) because we can use more precise clock that event queue here. | |
622 | * In such case we remove class from event queue first. | |
623 | */ | |
87990467 | 624 | static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, |
c9726d68 | 625 | int level, struct sk_buff *skb) |
87990467 | 626 | { |
0abf77e5 | 627 | int bytes = qdisc_pkt_len(skb); |
1da177e4 | 628 | enum htb_cmode old_mode; |
59e4220a | 629 | long diff; |
1da177e4 LT |
630 | |
631 | while (cl) { | |
03cc45c0 | 632 | diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer); |
1da177e4 | 633 | if (cl->level >= level) { |
87990467 SH |
634 | if (cl->level == level) |
635 | cl->xstats.lends++; | |
59e4220a | 636 | htb_accnt_tokens(cl, bytes, diff); |
1da177e4 LT |
637 | } else { |
638 | cl->xstats.borrows++; | |
87990467 | 639 | cl->tokens += diff; /* we moved t_c; update tokens */ |
1da177e4 | 640 | } |
59e4220a | 641 | htb_accnt_ctokens(cl, bytes, diff); |
1da177e4 | 642 | cl->t_c = q->now; |
1da177e4 | 643 | |
87990467 SH |
644 | old_mode = cl->cmode; |
645 | diff = 0; | |
646 | htb_change_class_mode(q, cl, &diff); | |
1da177e4 LT |
647 | if (old_mode != cl->cmode) { |
648 | if (old_mode != HTB_CAN_SEND) | |
3696f625 | 649 | htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level); |
1da177e4 | 650 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 651 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 652 | } |
1da177e4 | 653 | |
bfe0d029 ED |
654 | /* update basic stats except for leaves which are already updated */ |
655 | if (cl->level) | |
656 | bstats_update(&cl->bstats, skb); | |
657 | ||
1da177e4 LT |
658 | cl = cl->parent; |
659 | } | |
660 | } | |
661 | ||
662 | /** | |
663 | * htb_do_events - make mode changes to classes at the level | |
664 | * | |
fb983d45 | 665 | * Scans event queue for pending events and applies them. Returns time of |
1224736d | 666 | * next pending event (0 for no event in pq, q->now for too many events). |
fb983d45 | 667 | * Note: Applied are events whose have cl->pq_key <= q->now. |
1da177e4 | 668 | */ |
a73be040 JP |
669 | static psched_time_t htb_do_events(struct htb_sched *q, int level, |
670 | unsigned long start) | |
1da177e4 | 671 | { |
8f3ea33a | 672 | /* don't run for longer than 2 jiffies; 2 is used instead of |
cc7ec456 ED |
673 | * 1 to simplify things when jiffy is going to be incremented |
674 | * too soon | |
675 | */ | |
a73be040 | 676 | unsigned long stop_at = start + 2; |
8f3ea33a | 677 | while (time_before(jiffies, stop_at)) { |
1da177e4 LT |
678 | struct htb_class *cl; |
679 | long diff; | |
30bdbe39 AM |
680 | struct rb_node *p = rb_first(&q->wait_pq[level]); |
681 | ||
87990467 SH |
682 | if (!p) |
683 | return 0; | |
1da177e4 LT |
684 | |
685 | cl = rb_entry(p, struct htb_class, pq_node); | |
fb983d45 PM |
686 | if (cl->pq_key > q->now) |
687 | return cl->pq_key; | |
688 | ||
3696f625 | 689 | htb_safe_rb_erase(p, q->wait_pq + level); |
03cc45c0 | 690 | diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer); |
87990467 | 691 | htb_change_class_mode(q, cl, &diff); |
1da177e4 | 692 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 693 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 694 | } |
1224736d JP |
695 | |
696 | /* too much load - let's continue after a break for scheduling */ | |
e82181de | 697 | if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) { |
cc7ec456 | 698 | pr_warning("htb: too many events!\n"); |
e82181de JP |
699 | q->warned |= HTB_WARN_TOOMANYEVENTS; |
700 | } | |
1224736d JP |
701 | |
702 | return q->now; | |
1da177e4 LT |
703 | } |
704 | ||
705 | /* Returns class->node+prio from id-tree where classe's id is >= id. NULL | |
cc7ec456 ED |
706 | * is no such one exists. |
707 | */ | |
87990467 SH |
708 | static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n, |
709 | u32 id) | |
1da177e4 LT |
710 | { |
711 | struct rb_node *r = NULL; | |
712 | while (n) { | |
87990467 SH |
713 | struct htb_class *cl = |
714 | rb_entry(n, struct htb_class, node[prio]); | |
87990467 | 715 | |
f4c1f3e0 | 716 | if (id > cl->common.classid) { |
1da177e4 | 717 | n = n->rb_right; |
1b5c0077 | 718 | } else if (id < cl->common.classid) { |
1da177e4 LT |
719 | r = n; |
720 | n = n->rb_left; | |
1b5c0077 JP |
721 | } else { |
722 | return n; | |
1da177e4 LT |
723 | } |
724 | } | |
725 | return r; | |
726 | } | |
727 | ||
728 | /** | |
729 | * htb_lookup_leaf - returns next leaf class in DRR order | |
730 | * | |
731 | * Find leaf where current feed pointers points to. | |
732 | */ | |
87990467 SH |
733 | static struct htb_class *htb_lookup_leaf(struct rb_root *tree, int prio, |
734 | struct rb_node **pptr, u32 * pid) | |
1da177e4 LT |
735 | { |
736 | int i; | |
737 | struct { | |
738 | struct rb_node *root; | |
739 | struct rb_node **pptr; | |
740 | u32 *pid; | |
87990467 SH |
741 | } stk[TC_HTB_MAXDEPTH], *sp = stk; |
742 | ||
512bb43e | 743 | BUG_ON(!tree->rb_node); |
1da177e4 LT |
744 | sp->root = tree->rb_node; |
745 | sp->pptr = pptr; | |
746 | sp->pid = pid; | |
747 | ||
748 | for (i = 0; i < 65535; i++) { | |
87990467 | 749 | if (!*sp->pptr && *sp->pid) { |
10297b99 | 750 | /* ptr was invalidated but id is valid - try to recover |
cc7ec456 ED |
751 | * the original or next ptr |
752 | */ | |
87990467 SH |
753 | *sp->pptr = |
754 | htb_id_find_next_upper(prio, sp->root, *sp->pid); | |
1da177e4 | 755 | } |
87990467 | 756 | *sp->pid = 0; /* ptr is valid now so that remove this hint as it |
cc7ec456 ED |
757 | * can become out of date quickly |
758 | */ | |
87990467 | 759 | if (!*sp->pptr) { /* we are at right end; rewind & go up */ |
1da177e4 | 760 | *sp->pptr = sp->root; |
87990467 | 761 | while ((*sp->pptr)->rb_left) |
1da177e4 LT |
762 | *sp->pptr = (*sp->pptr)->rb_left; |
763 | if (sp > stk) { | |
764 | sp--; | |
512bb43e JP |
765 | if (!*sp->pptr) { |
766 | WARN_ON(1); | |
87990467 | 767 | return NULL; |
512bb43e | 768 | } |
87990467 | 769 | htb_next_rb_node(sp->pptr); |
1da177e4 LT |
770 | } |
771 | } else { | |
772 | struct htb_class *cl; | |
87990467 SH |
773 | cl = rb_entry(*sp->pptr, struct htb_class, node[prio]); |
774 | if (!cl->level) | |
1da177e4 LT |
775 | return cl; |
776 | (++sp)->root = cl->un.inner.feed[prio].rb_node; | |
87990467 SH |
777 | sp->pptr = cl->un.inner.ptr + prio; |
778 | sp->pid = cl->un.inner.last_ptr_id + prio; | |
1da177e4 LT |
779 | } |
780 | } | |
547b792c | 781 | WARN_ON(1); |
1da177e4 LT |
782 | return NULL; |
783 | } | |
784 | ||
785 | /* dequeues packet at given priority and level; call only if | |
cc7ec456 ED |
786 | * you are sure that there is active class at prio/level |
787 | */ | |
87990467 SH |
788 | static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, int prio, |
789 | int level) | |
1da177e4 LT |
790 | { |
791 | struct sk_buff *skb = NULL; | |
87990467 | 792 | struct htb_class *cl, *start; |
1da177e4 | 793 | /* look initial class up in the row */ |
87990467 SH |
794 | start = cl = htb_lookup_leaf(q->row[level] + prio, prio, |
795 | q->ptr[level] + prio, | |
796 | q->last_ptr_id[level] + prio); | |
797 | ||
1da177e4 LT |
798 | do { |
799 | next: | |
512bb43e | 800 | if (unlikely(!cl)) |
87990467 | 801 | return NULL; |
1da177e4 LT |
802 | |
803 | /* class can be empty - it is unlikely but can be true if leaf | |
cc7ec456 ED |
804 | * qdisc drops packets in enqueue routine or if someone used |
805 | * graft operation on the leaf since last dequeue; | |
806 | * simply deactivate and skip such class | |
807 | */ | |
1da177e4 LT |
808 | if (unlikely(cl->un.leaf.q->q.qlen == 0)) { |
809 | struct htb_class *next; | |
87990467 | 810 | htb_deactivate(q, cl); |
1da177e4 LT |
811 | |
812 | /* row/level might become empty */ | |
813 | if ((q->row_mask[level] & (1 << prio)) == 0) | |
87990467 | 814 | return NULL; |
1da177e4 | 815 | |
87990467 SH |
816 | next = htb_lookup_leaf(q->row[level] + prio, |
817 | prio, q->ptr[level] + prio, | |
818 | q->last_ptr_id[level] + prio); | |
819 | ||
820 | if (cl == start) /* fix start if we just deleted it */ | |
1da177e4 LT |
821 | start = next; |
822 | cl = next; | |
823 | goto next; | |
824 | } | |
87990467 SH |
825 | |
826 | skb = cl->un.leaf.q->dequeue(cl->un.leaf.q); | |
827 | if (likely(skb != NULL)) | |
1da177e4 | 828 | break; |
633fe66e | 829 | |
b00355db | 830 | qdisc_warn_nonwc("htb", cl->un.leaf.q); |
87990467 SH |
831 | htb_next_rb_node((level ? cl->parent->un.inner.ptr : q-> |
832 | ptr[0]) + prio); | |
833 | cl = htb_lookup_leaf(q->row[level] + prio, prio, | |
834 | q->ptr[level] + prio, | |
835 | q->last_ptr_id[level] + prio); | |
1da177e4 LT |
836 | |
837 | } while (cl != start); | |
838 | ||
839 | if (likely(skb != NULL)) { | |
0abf77e5 JK |
840 | cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb); |
841 | if (cl->un.leaf.deficit[level] < 0) { | |
c19f7a34 | 842 | cl->un.leaf.deficit[level] += cl->quantum; |
87990467 SH |
843 | htb_next_rb_node((level ? cl->parent->un.inner.ptr : q-> |
844 | ptr[0]) + prio); | |
1da177e4 LT |
845 | } |
846 | /* this used to be after charge_class but this constelation | |
cc7ec456 ED |
847 | * gives us slightly better performance |
848 | */ | |
1da177e4 | 849 | if (!cl->un.leaf.q->q.qlen) |
87990467 | 850 | htb_deactivate(q, cl); |
c9726d68 | 851 | htb_charge_class(q, cl, level, skb); |
1da177e4 LT |
852 | } |
853 | return skb; | |
854 | } | |
855 | ||
1da177e4 LT |
856 | static struct sk_buff *htb_dequeue(struct Qdisc *sch) |
857 | { | |
9190b3b3 | 858 | struct sk_buff *skb; |
1da177e4 LT |
859 | struct htb_sched *q = qdisc_priv(sch); |
860 | int level; | |
fb983d45 | 861 | psched_time_t next_event; |
a73be040 | 862 | unsigned long start_at; |
1da177e4 LT |
863 | |
864 | /* try to dequeue direct packets as high prio (!) to minimize cpu work */ | |
87990467 SH |
865 | skb = __skb_dequeue(&q->direct_queue); |
866 | if (skb != NULL) { | |
9190b3b3 ED |
867 | ok: |
868 | qdisc_bstats_update(sch, skb); | |
fd245a4a | 869 | qdisc_unthrottled(sch); |
1da177e4 LT |
870 | sch->q.qlen--; |
871 | return skb; | |
872 | } | |
873 | ||
87990467 SH |
874 | if (!sch->q.qlen) |
875 | goto fin; | |
3bebcda2 | 876 | q->now = psched_get_time(); |
a73be040 | 877 | start_at = jiffies; |
1da177e4 | 878 | |
fb983d45 | 879 | next_event = q->now + 5 * PSCHED_TICKS_PER_SEC; |
633fe66e | 880 | |
1da177e4 LT |
881 | for (level = 0; level < TC_HTB_MAXDEPTH; level++) { |
882 | /* common case optimization - skip event handler quickly */ | |
883 | int m; | |
fb983d45 PM |
884 | psched_time_t event; |
885 | ||
886 | if (q->now >= q->near_ev_cache[level]) { | |
a73be040 | 887 | event = htb_do_events(q, level, start_at); |
2e4b3b0e PM |
888 | if (!event) |
889 | event = q->now + PSCHED_TICKS_PER_SEC; | |
890 | q->near_ev_cache[level] = event; | |
1da177e4 | 891 | } else |
fb983d45 PM |
892 | event = q->near_ev_cache[level]; |
893 | ||
c0851347 | 894 | if (next_event > event) |
fb983d45 | 895 | next_event = event; |
87990467 | 896 | |
1da177e4 LT |
897 | m = ~q->row_mask[level]; |
898 | while (m != (int)(-1)) { | |
87990467 | 899 | int prio = ffz(m); |
cc7ec456 | 900 | |
1da177e4 | 901 | m |= 1 << prio; |
87990467 | 902 | skb = htb_dequeue_tree(q, prio, level); |
9190b3b3 ED |
903 | if (likely(skb != NULL)) |
904 | goto ok; | |
1da177e4 LT |
905 | } |
906 | } | |
fb983d45 | 907 | sch->qstats.overlimits++; |
1224736d JP |
908 | if (likely(next_event > q->now)) |
909 | qdisc_watchdog_schedule(&q->watchdog, next_event); | |
910 | else | |
911 | schedule_work(&q->work); | |
1da177e4 | 912 | fin: |
1da177e4 LT |
913 | return skb; |
914 | } | |
915 | ||
916 | /* try to drop from each class (by prio) until one succeed */ | |
87990467 | 917 | static unsigned int htb_drop(struct Qdisc *sch) |
1da177e4 LT |
918 | { |
919 | struct htb_sched *q = qdisc_priv(sch); | |
920 | int prio; | |
921 | ||
922 | for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) { | |
923 | struct list_head *p; | |
87990467 | 924 | list_for_each(p, q->drops + prio) { |
1da177e4 LT |
925 | struct htb_class *cl = list_entry(p, struct htb_class, |
926 | un.leaf.drop_list); | |
927 | unsigned int len; | |
87990467 SH |
928 | if (cl->un.leaf.q->ops->drop && |
929 | (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) { | |
1da177e4 LT |
930 | sch->q.qlen--; |
931 | if (!cl->un.leaf.q->q.qlen) | |
87990467 | 932 | htb_deactivate(q, cl); |
1da177e4 LT |
933 | return len; |
934 | } | |
935 | } | |
936 | } | |
937 | return 0; | |
938 | } | |
939 | ||
940 | /* reset all classes */ | |
941 | /* always caled under BH & queue lock */ | |
87990467 | 942 | static void htb_reset(struct Qdisc *sch) |
1da177e4 LT |
943 | { |
944 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 PM |
945 | struct htb_class *cl; |
946 | struct hlist_node *n; | |
947 | unsigned int i; | |
0cef296d | 948 | |
f4c1f3e0 PM |
949 | for (i = 0; i < q->clhash.hashsize; i++) { |
950 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { | |
1da177e4 | 951 | if (cl->level) |
87990467 | 952 | memset(&cl->un.inner, 0, sizeof(cl->un.inner)); |
1da177e4 | 953 | else { |
87990467 | 954 | if (cl->un.leaf.q) |
1da177e4 LT |
955 | qdisc_reset(cl->un.leaf.q); |
956 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); | |
957 | } | |
958 | cl->prio_activity = 0; | |
959 | cl->cmode = HTB_CAN_SEND; | |
1da177e4 LT |
960 | |
961 | } | |
962 | } | |
fb983d45 | 963 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
964 | __skb_queue_purge(&q->direct_queue); |
965 | sch->q.qlen = 0; | |
87990467 SH |
966 | memset(q->row, 0, sizeof(q->row)); |
967 | memset(q->row_mask, 0, sizeof(q->row_mask)); | |
968 | memset(q->wait_pq, 0, sizeof(q->wait_pq)); | |
969 | memset(q->ptr, 0, sizeof(q->ptr)); | |
1da177e4 | 970 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 971 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 LT |
972 | } |
973 | ||
27a3421e PM |
974 | static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = { |
975 | [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) }, | |
976 | [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) }, | |
977 | [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
978 | [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
979 | }; | |
980 | ||
1224736d JP |
981 | static void htb_work_func(struct work_struct *work) |
982 | { | |
983 | struct htb_sched *q = container_of(work, struct htb_sched, work); | |
984 | struct Qdisc *sch = q->watchdog.qdisc; | |
985 | ||
986 | __netif_schedule(qdisc_root(sch)); | |
987 | } | |
988 | ||
1e90474c | 989 | static int htb_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
990 | { |
991 | struct htb_sched *q = qdisc_priv(sch); | |
1e90474c | 992 | struct nlattr *tb[TCA_HTB_INIT + 1]; |
1da177e4 | 993 | struct tc_htb_glob *gopt; |
cee63723 | 994 | int err; |
1da177e4 | 995 | int i; |
cee63723 PM |
996 | |
997 | if (!opt) | |
998 | return -EINVAL; | |
999 | ||
27a3421e | 1000 | err = nla_parse_nested(tb, TCA_HTB_INIT, opt, htb_policy); |
cee63723 PM |
1001 | if (err < 0) |
1002 | return err; | |
1003 | ||
27a3421e | 1004 | if (tb[TCA_HTB_INIT] == NULL) { |
cc7ec456 | 1005 | pr_err("HTB: hey probably you have bad tc tool ?\n"); |
1da177e4 LT |
1006 | return -EINVAL; |
1007 | } | |
1e90474c | 1008 | gopt = nla_data(tb[TCA_HTB_INIT]); |
1da177e4 | 1009 | if (gopt->version != HTB_VER >> 16) { |
cc7ec456 | 1010 | pr_err("HTB: need tc/htb version %d (minor is %d), you have %d\n", |
87990467 | 1011 | HTB_VER >> 16, HTB_VER & 0xffff, gopt->version); |
1da177e4 LT |
1012 | return -EINVAL; |
1013 | } | |
1da177e4 | 1014 | |
f4c1f3e0 PM |
1015 | err = qdisc_class_hash_init(&q->clhash); |
1016 | if (err < 0) | |
1017 | return err; | |
1da177e4 | 1018 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 1019 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 | 1020 | |
fb983d45 | 1021 | qdisc_watchdog_init(&q->watchdog, sch); |
1224736d | 1022 | INIT_WORK(&q->work, htb_work_func); |
1da177e4 LT |
1023 | skb_queue_head_init(&q->direct_queue); |
1024 | ||
5ce2d488 | 1025 | q->direct_qlen = qdisc_dev(sch)->tx_queue_len; |
87990467 | 1026 | if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */ |
1da177e4 | 1027 | q->direct_qlen = 2; |
1da177e4 | 1028 | |
1da177e4 LT |
1029 | if ((q->rate2quantum = gopt->rate2quantum) < 1) |
1030 | q->rate2quantum = 1; | |
1031 | q->defcls = gopt->defcls; | |
1032 | ||
1033 | return 0; | |
1034 | } | |
1035 | ||
1036 | static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) | |
1037 | { | |
102396ae | 1038 | spinlock_t *root_lock = qdisc_root_sleeping_lock(sch); |
1da177e4 | 1039 | struct htb_sched *q = qdisc_priv(sch); |
4b3550ef | 1040 | struct nlattr *nest; |
1da177e4 | 1041 | struct tc_htb_glob gopt; |
4b3550ef | 1042 | |
7698b4fc | 1043 | spin_lock_bh(root_lock); |
1da177e4 | 1044 | |
4b3550ef | 1045 | gopt.direct_pkts = q->direct_pkts; |
1da177e4 LT |
1046 | gopt.version = HTB_VER; |
1047 | gopt.rate2quantum = q->rate2quantum; | |
1048 | gopt.defcls = q->defcls; | |
3bf72957 | 1049 | gopt.debug = 0; |
4b3550ef PM |
1050 | |
1051 | nest = nla_nest_start(skb, TCA_OPTIONS); | |
1052 | if (nest == NULL) | |
1053 | goto nla_put_failure; | |
1e90474c | 1054 | NLA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt); |
4b3550ef PM |
1055 | nla_nest_end(skb, nest); |
1056 | ||
7698b4fc | 1057 | spin_unlock_bh(root_lock); |
1da177e4 | 1058 | return skb->len; |
4b3550ef | 1059 | |
1e90474c | 1060 | nla_put_failure: |
7698b4fc | 1061 | spin_unlock_bh(root_lock); |
4b3550ef | 1062 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1063 | return -1; |
1064 | } | |
1065 | ||
1066 | static int htb_dump_class(struct Qdisc *sch, unsigned long arg, | |
87990467 | 1067 | struct sk_buff *skb, struct tcmsg *tcm) |
1da177e4 | 1068 | { |
87990467 | 1069 | struct htb_class *cl = (struct htb_class *)arg; |
102396ae | 1070 | spinlock_t *root_lock = qdisc_root_sleeping_lock(sch); |
4b3550ef | 1071 | struct nlattr *nest; |
1da177e4 LT |
1072 | struct tc_htb_opt opt; |
1073 | ||
7698b4fc | 1074 | spin_lock_bh(root_lock); |
f4c1f3e0 PM |
1075 | tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT; |
1076 | tcm->tcm_handle = cl->common.classid; | |
1da177e4 LT |
1077 | if (!cl->level && cl->un.leaf.q) |
1078 | tcm->tcm_info = cl->un.leaf.q->handle; | |
1079 | ||
4b3550ef PM |
1080 | nest = nla_nest_start(skb, TCA_OPTIONS); |
1081 | if (nest == NULL) | |
1082 | goto nla_put_failure; | |
1da177e4 | 1083 | |
87990467 | 1084 | memset(&opt, 0, sizeof(opt)); |
1da177e4 | 1085 | |
87990467 SH |
1086 | opt.rate = cl->rate->rate; |
1087 | opt.buffer = cl->buffer; | |
1088 | opt.ceil = cl->ceil->rate; | |
1089 | opt.cbuffer = cl->cbuffer; | |
c19f7a34 JP |
1090 | opt.quantum = cl->quantum; |
1091 | opt.prio = cl->prio; | |
87990467 | 1092 | opt.level = cl->level; |
1e90474c | 1093 | NLA_PUT(skb, TCA_HTB_PARMS, sizeof(opt), &opt); |
4b3550ef PM |
1094 | |
1095 | nla_nest_end(skb, nest); | |
7698b4fc | 1096 | spin_unlock_bh(root_lock); |
1da177e4 | 1097 | return skb->len; |
4b3550ef | 1098 | |
1e90474c | 1099 | nla_put_failure: |
7698b4fc | 1100 | spin_unlock_bh(root_lock); |
4b3550ef | 1101 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
1102 | return -1; |
1103 | } | |
1104 | ||
1105 | static int | |
87990467 | 1106 | htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d) |
1da177e4 | 1107 | { |
87990467 | 1108 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 | 1109 | |
1da177e4 LT |
1110 | if (!cl->level && cl->un.leaf.q) |
1111 | cl->qstats.qlen = cl->un.leaf.q->q.qlen; | |
1112 | cl->xstats.tokens = cl->tokens; | |
1113 | cl->xstats.ctokens = cl->ctokens; | |
1114 | ||
1115 | if (gnet_stats_copy_basic(d, &cl->bstats) < 0 || | |
d250a5f9 | 1116 | gnet_stats_copy_rate_est(d, NULL, &cl->rate_est) < 0 || |
1da177e4 LT |
1117 | gnet_stats_copy_queue(d, &cl->qstats) < 0) |
1118 | return -1; | |
1119 | ||
1120 | return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); | |
1121 | } | |
1122 | ||
1123 | static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
87990467 | 1124 | struct Qdisc **old) |
1da177e4 | 1125 | { |
87990467 | 1126 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 | 1127 | |
5b9a9ccf PM |
1128 | if (cl->level) |
1129 | return -EINVAL; | |
1130 | if (new == NULL && | |
3511c913 | 1131 | (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
5b9a9ccf PM |
1132 | cl->common.classid)) == NULL) |
1133 | return -ENOBUFS; | |
1134 | ||
1135 | sch_tree_lock(sch); | |
1136 | *old = cl->un.leaf.q; | |
1137 | cl->un.leaf.q = new; | |
1138 | if (*old != NULL) { | |
1139 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); | |
1140 | qdisc_reset(*old); | |
1da177e4 | 1141 | } |
5b9a9ccf PM |
1142 | sch_tree_unlock(sch); |
1143 | return 0; | |
1da177e4 LT |
1144 | } |
1145 | ||
87990467 | 1146 | static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg) |
1da177e4 | 1147 | { |
87990467 | 1148 | struct htb_class *cl = (struct htb_class *)arg; |
5b9a9ccf | 1149 | return !cl->level ? cl->un.leaf.q : NULL; |
1da177e4 LT |
1150 | } |
1151 | ||
256d61b8 PM |
1152 | static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg) |
1153 | { | |
1154 | struct htb_class *cl = (struct htb_class *)arg; | |
1155 | ||
1156 | if (cl->un.leaf.q->q.qlen == 0) | |
1157 | htb_deactivate(qdisc_priv(sch), cl); | |
1158 | } | |
1159 | ||
1da177e4 LT |
1160 | static unsigned long htb_get(struct Qdisc *sch, u32 classid) |
1161 | { | |
87990467 SH |
1162 | struct htb_class *cl = htb_find(classid, sch); |
1163 | if (cl) | |
1da177e4 LT |
1164 | cl->refcnt++; |
1165 | return (unsigned long)cl; | |
1166 | } | |
1167 | ||
160d5e10 JP |
1168 | static inline int htb_parent_last_child(struct htb_class *cl) |
1169 | { | |
1170 | if (!cl->parent) | |
1171 | /* the root class */ | |
1172 | return 0; | |
42077599 | 1173 | if (cl->parent->children > 1) |
160d5e10 JP |
1174 | /* not the last child */ |
1175 | return 0; | |
160d5e10 JP |
1176 | return 1; |
1177 | } | |
1178 | ||
3ba08b00 JP |
1179 | static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl, |
1180 | struct Qdisc *new_q) | |
160d5e10 JP |
1181 | { |
1182 | struct htb_class *parent = cl->parent; | |
1183 | ||
547b792c | 1184 | WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity); |
160d5e10 | 1185 | |
3ba08b00 JP |
1186 | if (parent->cmode != HTB_CAN_SEND) |
1187 | htb_safe_rb_erase(&parent->pq_node, q->wait_pq + parent->level); | |
1188 | ||
160d5e10 JP |
1189 | parent->level = 0; |
1190 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1191 | INIT_LIST_HEAD(&parent->un.leaf.drop_list); | |
1192 | parent->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
160d5e10 JP |
1193 | parent->tokens = parent->buffer; |
1194 | parent->ctokens = parent->cbuffer; | |
3bebcda2 | 1195 | parent->t_c = psched_get_time(); |
160d5e10 JP |
1196 | parent->cmode = HTB_CAN_SEND; |
1197 | } | |
1198 | ||
87990467 | 1199 | static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl) |
1da177e4 | 1200 | { |
1da177e4 | 1201 | if (!cl->level) { |
547b792c | 1202 | WARN_ON(!cl->un.leaf.q); |
1da177e4 LT |
1203 | qdisc_destroy(cl->un.leaf.q); |
1204 | } | |
ee39e10c | 1205 | gen_kill_estimator(&cl->bstats, &cl->rate_est); |
1da177e4 LT |
1206 | qdisc_put_rtab(cl->rate); |
1207 | qdisc_put_rtab(cl->ceil); | |
87990467 | 1208 | |
ff31ab56 | 1209 | tcf_destroy_chain(&cl->filter_list); |
1da177e4 LT |
1210 | kfree(cl); |
1211 | } | |
1212 | ||
87990467 | 1213 | static void htb_destroy(struct Qdisc *sch) |
1da177e4 LT |
1214 | { |
1215 | struct htb_sched *q = qdisc_priv(sch); | |
fbd8f137 PM |
1216 | struct hlist_node *n, *next; |
1217 | struct htb_class *cl; | |
1218 | unsigned int i; | |
1da177e4 | 1219 | |
1224736d | 1220 | cancel_work_sync(&q->work); |
fb983d45 | 1221 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 | 1222 | /* This line used to be after htb_destroy_class call below |
cc7ec456 ED |
1223 | * and surprisingly it worked in 2.4. But it must precede it |
1224 | * because filter need its target class alive to be able to call | |
1225 | * unbind_filter on it (without Oops). | |
1226 | */ | |
ff31ab56 | 1227 | tcf_destroy_chain(&q->filter_list); |
87990467 | 1228 | |
f4c1f3e0 PM |
1229 | for (i = 0; i < q->clhash.hashsize; i++) { |
1230 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) | |
fbd8f137 PM |
1231 | tcf_destroy_chain(&cl->filter_list); |
1232 | } | |
f4c1f3e0 PM |
1233 | for (i = 0; i < q->clhash.hashsize; i++) { |
1234 | hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i], | |
1235 | common.hnode) | |
fbd8f137 PM |
1236 | htb_destroy_class(sch, cl); |
1237 | } | |
f4c1f3e0 | 1238 | qdisc_class_hash_destroy(&q->clhash); |
1da177e4 LT |
1239 | __skb_queue_purge(&q->direct_queue); |
1240 | } | |
1241 | ||
1242 | static int htb_delete(struct Qdisc *sch, unsigned long arg) | |
1243 | { | |
1244 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1245 | struct htb_class *cl = (struct htb_class *)arg; |
256d61b8 | 1246 | unsigned int qlen; |
160d5e10 JP |
1247 | struct Qdisc *new_q = NULL; |
1248 | int last_child = 0; | |
1da177e4 LT |
1249 | |
1250 | // TODO: why don't allow to delete subtree ? references ? does | |
1251 | // tc subsys quarantee us that in htb_destroy it holds no class | |
1252 | // refs so that we can remove children safely there ? | |
42077599 | 1253 | if (cl->children || cl->filter_cnt) |
1da177e4 | 1254 | return -EBUSY; |
87990467 | 1255 | |
160d5e10 | 1256 | if (!cl->level && htb_parent_last_child(cl)) { |
3511c913 | 1257 | new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
bb949fbd | 1258 | cl->parent->common.classid); |
160d5e10 JP |
1259 | last_child = 1; |
1260 | } | |
1261 | ||
1da177e4 | 1262 | sch_tree_lock(sch); |
87990467 | 1263 | |
814a175e | 1264 | if (!cl->level) { |
256d61b8 | 1265 | qlen = cl->un.leaf.q->q.qlen; |
814a175e | 1266 | qdisc_reset(cl->un.leaf.q); |
256d61b8 | 1267 | qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen); |
814a175e PM |
1268 | } |
1269 | ||
f4c1f3e0 PM |
1270 | /* delete from hash and active; remainder in destroy_class */ |
1271 | qdisc_class_hash_remove(&q->clhash, &cl->common); | |
26b284de JP |
1272 | if (cl->parent) |
1273 | cl->parent->children--; | |
c38c83cb | 1274 | |
1da177e4 | 1275 | if (cl->prio_activity) |
87990467 | 1276 | htb_deactivate(q, cl); |
1da177e4 | 1277 | |
fbd8f137 PM |
1278 | if (cl->cmode != HTB_CAN_SEND) |
1279 | htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level); | |
1280 | ||
160d5e10 | 1281 | if (last_child) |
3ba08b00 | 1282 | htb_parent_to_leaf(q, cl, new_q); |
160d5e10 | 1283 | |
7cd0a638 JP |
1284 | BUG_ON(--cl->refcnt == 0); |
1285 | /* | |
1286 | * This shouldn't happen: we "hold" one cops->get() when called | |
1287 | * from tc_ctl_tclass; the destroy method is done from cops->put(). | |
1288 | */ | |
1da177e4 LT |
1289 | |
1290 | sch_tree_unlock(sch); | |
1291 | return 0; | |
1292 | } | |
1293 | ||
1294 | static void htb_put(struct Qdisc *sch, unsigned long arg) | |
1295 | { | |
87990467 | 1296 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1297 | |
1298 | if (--cl->refcnt == 0) | |
87990467 | 1299 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1300 | } |
1301 | ||
87990467 | 1302 | static int htb_change_class(struct Qdisc *sch, u32 classid, |
1e90474c | 1303 | u32 parentid, struct nlattr **tca, |
87990467 | 1304 | unsigned long *arg) |
1da177e4 LT |
1305 | { |
1306 | int err = -EINVAL; | |
1307 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1308 | struct htb_class *cl = (struct htb_class *)*arg, *parent; |
1e90474c | 1309 | struct nlattr *opt = tca[TCA_OPTIONS]; |
1da177e4 | 1310 | struct qdisc_rate_table *rtab = NULL, *ctab = NULL; |
e18434c4 | 1311 | struct nlattr *tb[__TCA_HTB_MAX]; |
1da177e4 LT |
1312 | struct tc_htb_opt *hopt; |
1313 | ||
1314 | /* extract all subattrs from opt attr */ | |
cee63723 PM |
1315 | if (!opt) |
1316 | goto failure; | |
1317 | ||
e18434c4 | 1318 | err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy); |
cee63723 PM |
1319 | if (err < 0) |
1320 | goto failure; | |
1321 | ||
1322 | err = -EINVAL; | |
27a3421e | 1323 | if (tb[TCA_HTB_PARMS] == NULL) |
1da177e4 | 1324 | goto failure; |
1da177e4 | 1325 | |
87990467 SH |
1326 | parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch); |
1327 | ||
1e90474c | 1328 | hopt = nla_data(tb[TCA_HTB_PARMS]); |
3bf72957 | 1329 | |
1e90474c PM |
1330 | rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB]); |
1331 | ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB]); | |
87990467 SH |
1332 | if (!rtab || !ctab) |
1333 | goto failure; | |
1da177e4 | 1334 | |
87990467 | 1335 | if (!cl) { /* new class */ |
1da177e4 | 1336 | struct Qdisc *new_q; |
3696f625 | 1337 | int prio; |
ee39e10c | 1338 | struct { |
1e90474c | 1339 | struct nlattr nla; |
ee39e10c PM |
1340 | struct gnet_estimator opt; |
1341 | } est = { | |
1e90474c PM |
1342 | .nla = { |
1343 | .nla_len = nla_attr_size(sizeof(est.opt)), | |
1344 | .nla_type = TCA_RATE, | |
ee39e10c PM |
1345 | }, |
1346 | .opt = { | |
1347 | /* 4s interval, 16s averaging constant */ | |
1348 | .interval = 2, | |
1349 | .ewma_log = 2, | |
1350 | }, | |
1351 | }; | |
3696f625 | 1352 | |
1da177e4 | 1353 | /* check for valid classid */ |
f64f9e71 JP |
1354 | if (!classid || TC_H_MAJ(classid ^ sch->handle) || |
1355 | htb_find(classid, sch)) | |
1da177e4 LT |
1356 | goto failure; |
1357 | ||
1358 | /* check maximal depth */ | |
1359 | if (parent && parent->parent && parent->parent->level < 2) { | |
cc7ec456 | 1360 | pr_err("htb: tree is too deep\n"); |
1da177e4 LT |
1361 | goto failure; |
1362 | } | |
1363 | err = -ENOBUFS; | |
cc7ec456 ED |
1364 | cl = kzalloc(sizeof(*cl), GFP_KERNEL); |
1365 | if (!cl) | |
1da177e4 | 1366 | goto failure; |
87990467 | 1367 | |
71bcb09a SH |
1368 | err = gen_new_estimator(&cl->bstats, &cl->rate_est, |
1369 | qdisc_root_sleeping_lock(sch), | |
1370 | tca[TCA_RATE] ? : &est.nla); | |
1371 | if (err) { | |
1372 | kfree(cl); | |
1373 | goto failure; | |
1374 | } | |
1375 | ||
1da177e4 | 1376 | cl->refcnt = 1; |
42077599 | 1377 | cl->children = 0; |
1da177e4 | 1378 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); |
3696f625 SH |
1379 | RB_CLEAR_NODE(&cl->pq_node); |
1380 | ||
1381 | for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) | |
1382 | RB_CLEAR_NODE(&cl->node[prio]); | |
1da177e4 LT |
1383 | |
1384 | /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL) | |
cc7ec456 ED |
1385 | * so that can't be used inside of sch_tree_lock |
1386 | * -- thanks to Karlis Peisenieks | |
1387 | */ | |
3511c913 | 1388 | new_q = qdisc_create_dflt(sch->dev_queue, |
bb949fbd | 1389 | &pfifo_qdisc_ops, classid); |
1da177e4 LT |
1390 | sch_tree_lock(sch); |
1391 | if (parent && !parent->level) { | |
256d61b8 PM |
1392 | unsigned int qlen = parent->un.leaf.q->q.qlen; |
1393 | ||
1da177e4 | 1394 | /* turn parent into inner node */ |
256d61b8 PM |
1395 | qdisc_reset(parent->un.leaf.q); |
1396 | qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen); | |
87990467 SH |
1397 | qdisc_destroy(parent->un.leaf.q); |
1398 | if (parent->prio_activity) | |
1399 | htb_deactivate(q, parent); | |
1da177e4 LT |
1400 | |
1401 | /* remove from evt list because of level change */ | |
1402 | if (parent->cmode != HTB_CAN_SEND) { | |
3696f625 | 1403 | htb_safe_rb_erase(&parent->pq_node, q->wait_pq); |
1da177e4 LT |
1404 | parent->cmode = HTB_CAN_SEND; |
1405 | } | |
1406 | parent->level = (parent->parent ? parent->parent->level | |
87990467 SH |
1407 | : TC_HTB_MAXDEPTH) - 1; |
1408 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1da177e4 LT |
1409 | } |
1410 | /* leaf (we) needs elementary qdisc */ | |
1411 | cl->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
1412 | ||
f4c1f3e0 | 1413 | cl->common.classid = classid; |
87990467 | 1414 | cl->parent = parent; |
1da177e4 LT |
1415 | |
1416 | /* set class to be in HTB_CAN_SEND state */ | |
1417 | cl->tokens = hopt->buffer; | |
1418 | cl->ctokens = hopt->cbuffer; | |
00c04af9 | 1419 | cl->mbuffer = 60 * PSCHED_TICKS_PER_SEC; /* 1min */ |
3bebcda2 | 1420 | cl->t_c = psched_get_time(); |
1da177e4 LT |
1421 | cl->cmode = HTB_CAN_SEND; |
1422 | ||
1423 | /* attach to the hash list and parent's family */ | |
f4c1f3e0 | 1424 | qdisc_class_hash_insert(&q->clhash, &cl->common); |
42077599 PM |
1425 | if (parent) |
1426 | parent->children++; | |
ee39e10c | 1427 | } else { |
71bcb09a SH |
1428 | if (tca[TCA_RATE]) { |
1429 | err = gen_replace_estimator(&cl->bstats, &cl->rate_est, | |
1430 | qdisc_root_sleeping_lock(sch), | |
1431 | tca[TCA_RATE]); | |
1432 | if (err) | |
1433 | return err; | |
1434 | } | |
87990467 | 1435 | sch_tree_lock(sch); |
ee39e10c | 1436 | } |
1da177e4 LT |
1437 | |
1438 | /* it used to be a nasty bug here, we have to check that node | |
cc7ec456 ED |
1439 | * is really leaf before changing cl->un.leaf ! |
1440 | */ | |
1da177e4 | 1441 | if (!cl->level) { |
c19f7a34 JP |
1442 | cl->quantum = rtab->rate.rate / q->rate2quantum; |
1443 | if (!hopt->quantum && cl->quantum < 1000) { | |
cc7ec456 | 1444 | pr_warning( |
87990467 | 1445 | "HTB: quantum of class %X is small. Consider r2q change.\n", |
f4c1f3e0 | 1446 | cl->common.classid); |
c19f7a34 | 1447 | cl->quantum = 1000; |
1da177e4 | 1448 | } |
c19f7a34 | 1449 | if (!hopt->quantum && cl->quantum > 200000) { |
cc7ec456 | 1450 | pr_warning( |
87990467 | 1451 | "HTB: quantum of class %X is big. Consider r2q change.\n", |
f4c1f3e0 | 1452 | cl->common.classid); |
c19f7a34 | 1453 | cl->quantum = 200000; |
1da177e4 LT |
1454 | } |
1455 | if (hopt->quantum) | |
c19f7a34 JP |
1456 | cl->quantum = hopt->quantum; |
1457 | if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO) | |
1458 | cl->prio = TC_HTB_NUMPRIO - 1; | |
1da177e4 LT |
1459 | } |
1460 | ||
1461 | cl->buffer = hopt->buffer; | |
1462 | cl->cbuffer = hopt->cbuffer; | |
87990467 SH |
1463 | if (cl->rate) |
1464 | qdisc_put_rtab(cl->rate); | |
1465 | cl->rate = rtab; | |
1466 | if (cl->ceil) | |
1467 | qdisc_put_rtab(cl->ceil); | |
1468 | cl->ceil = ctab; | |
1da177e4 LT |
1469 | sch_tree_unlock(sch); |
1470 | ||
f4c1f3e0 PM |
1471 | qdisc_class_hash_grow(sch, &q->clhash); |
1472 | ||
1da177e4 LT |
1473 | *arg = (unsigned long)cl; |
1474 | return 0; | |
1475 | ||
1476 | failure: | |
87990467 SH |
1477 | if (rtab) |
1478 | qdisc_put_rtab(rtab); | |
1479 | if (ctab) | |
1480 | qdisc_put_rtab(ctab); | |
1da177e4 LT |
1481 | return err; |
1482 | } | |
1483 | ||
1484 | static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg) | |
1485 | { | |
1486 | struct htb_sched *q = qdisc_priv(sch); | |
1487 | struct htb_class *cl = (struct htb_class *)arg; | |
1488 | struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list; | |
3bf72957 | 1489 | |
1da177e4 LT |
1490 | return fl; |
1491 | } | |
1492 | ||
1493 | static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, | |
87990467 | 1494 | u32 classid) |
1da177e4 | 1495 | { |
87990467 | 1496 | struct htb_class *cl = htb_find(classid, sch); |
3bf72957 | 1497 | |
1da177e4 | 1498 | /*if (cl && !cl->level) return 0; |
cc7ec456 ED |
1499 | * The line above used to be there to prevent attaching filters to |
1500 | * leaves. But at least tc_index filter uses this just to get class | |
1501 | * for other reasons so that we have to allow for it. | |
1502 | * ---- | |
1503 | * 19.6.2002 As Werner explained it is ok - bind filter is just | |
1504 | * another way to "lock" the class - unlike "get" this lock can | |
1505 | * be broken by class during destroy IIUC. | |
1da177e4 | 1506 | */ |
87990467 SH |
1507 | if (cl) |
1508 | cl->filter_cnt++; | |
1da177e4 LT |
1509 | return (unsigned long)cl; |
1510 | } | |
1511 | ||
1512 | static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) | |
1513 | { | |
1da177e4 | 1514 | struct htb_class *cl = (struct htb_class *)arg; |
3bf72957 | 1515 | |
87990467 SH |
1516 | if (cl) |
1517 | cl->filter_cnt--; | |
1da177e4 LT |
1518 | } |
1519 | ||
1520 | static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
1521 | { | |
1522 | struct htb_sched *q = qdisc_priv(sch); | |
f4c1f3e0 PM |
1523 | struct htb_class *cl; |
1524 | struct hlist_node *n; | |
1525 | unsigned int i; | |
1da177e4 LT |
1526 | |
1527 | if (arg->stop) | |
1528 | return; | |
1529 | ||
f4c1f3e0 PM |
1530 | for (i = 0; i < q->clhash.hashsize; i++) { |
1531 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { | |
1da177e4 LT |
1532 | if (arg->count < arg->skip) { |
1533 | arg->count++; | |
1534 | continue; | |
1535 | } | |
1536 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
1537 | arg->stop = 1; | |
1538 | return; | |
1539 | } | |
1540 | arg->count++; | |
1541 | } | |
1542 | } | |
1543 | } | |
1544 | ||
20fea08b | 1545 | static const struct Qdisc_class_ops htb_class_ops = { |
1da177e4 LT |
1546 | .graft = htb_graft, |
1547 | .leaf = htb_leaf, | |
256d61b8 | 1548 | .qlen_notify = htb_qlen_notify, |
1da177e4 LT |
1549 | .get = htb_get, |
1550 | .put = htb_put, | |
1551 | .change = htb_change_class, | |
1552 | .delete = htb_delete, | |
1553 | .walk = htb_walk, | |
1554 | .tcf_chain = htb_find_tcf, | |
1555 | .bind_tcf = htb_bind_filter, | |
1556 | .unbind_tcf = htb_unbind_filter, | |
1557 | .dump = htb_dump_class, | |
1558 | .dump_stats = htb_dump_class_stats, | |
1559 | }; | |
1560 | ||
20fea08b | 1561 | static struct Qdisc_ops htb_qdisc_ops __read_mostly = { |
1da177e4 LT |
1562 | .cl_ops = &htb_class_ops, |
1563 | .id = "htb", | |
1564 | .priv_size = sizeof(struct htb_sched), | |
1565 | .enqueue = htb_enqueue, | |
1566 | .dequeue = htb_dequeue, | |
77be155c | 1567 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1568 | .drop = htb_drop, |
1569 | .init = htb_init, | |
1570 | .reset = htb_reset, | |
1571 | .destroy = htb_destroy, | |
1da177e4 LT |
1572 | .dump = htb_dump, |
1573 | .owner = THIS_MODULE, | |
1574 | }; | |
1575 | ||
1576 | static int __init htb_module_init(void) | |
1577 | { | |
87990467 | 1578 | return register_qdisc(&htb_qdisc_ops); |
1da177e4 | 1579 | } |
87990467 | 1580 | static void __exit htb_module_exit(void) |
1da177e4 | 1581 | { |
87990467 | 1582 | unregister_qdisc(&htb_qdisc_ops); |
1da177e4 | 1583 | } |
87990467 | 1584 | |
1da177e4 LT |
1585 | module_init(htb_module_init) |
1586 | module_exit(htb_module_exit) | |
1587 | MODULE_LICENSE("GPL"); |