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. | |
27 | * | |
28 | * $Id: sch_htb.c,v 1.25 2003/12/07 11:08:25 devik Exp devik $ | |
29 | */ | |
1da177e4 LT |
30 | #include <linux/module.h> |
31 | #include <asm/uaccess.h> | |
32 | #include <asm/system.h> | |
33 | #include <linux/bitops.h> | |
34 | #include <linux/types.h> | |
35 | #include <linux/kernel.h> | |
1da177e4 LT |
36 | #include <linux/string.h> |
37 | #include <linux/mm.h> | |
38 | #include <linux/socket.h> | |
39 | #include <linux/sockios.h> | |
40 | #include <linux/in.h> | |
41 | #include <linux/errno.h> | |
42 | #include <linux/interrupt.h> | |
43 | #include <linux/if_ether.h> | |
44 | #include <linux/inet.h> | |
45 | #include <linux/netdevice.h> | |
46 | #include <linux/etherdevice.h> | |
47 | #include <linux/notifier.h> | |
48 | #include <net/ip.h> | |
49 | #include <net/route.h> | |
50 | #include <linux/skbuff.h> | |
51 | #include <linux/list.h> | |
52 | #include <linux/compiler.h> | |
dc5fc579 | 53 | #include <net/netlink.h> |
1da177e4 LT |
54 | #include <net/sock.h> |
55 | #include <net/pkt_sched.h> | |
56 | #include <linux/rbtree.h> | |
57 | ||
58 | /* HTB algorithm. | |
59 | Author: devik@cdi.cz | |
60 | ======================================================================== | |
61 | HTB is like TBF with multiple classes. It is also similar to CBQ because | |
10297b99 | 62 | it allows to assign priority to each class in hierarchy. |
1da177e4 LT |
63 | In fact it is another implementation of Floyd's formal sharing. |
64 | ||
65 | Levels: | |
10297b99 | 66 | Each class is assigned level. Leaf has ALWAYS level 0 and root |
1da177e4 LT |
67 | classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level |
68 | one less than their parent. | |
69 | */ | |
70 | ||
87990467 SH |
71 | #define HTB_HSIZE 16 /* classid hash size */ |
72 | #define HTB_EWMAC 2 /* rate average over HTB_EWMAC*HTB_HSIZE sec */ | |
73 | #define HTB_RATECM 1 /* whether to use rate computer */ | |
74 | #define HTB_HYSTERESIS 1 /* whether to use mode hysteresis for speedup */ | |
75 | #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */ | |
1da177e4 LT |
76 | |
77 | #if HTB_VER >> 16 != TC_HTB_PROTOVER | |
78 | #error "Mismatched sch_htb.c and pkt_sch.h" | |
79 | #endif | |
80 | ||
1da177e4 LT |
81 | /* used internaly to keep status of single class */ |
82 | enum htb_cmode { | |
87990467 SH |
83 | HTB_CANT_SEND, /* class can't send and can't borrow */ |
84 | HTB_MAY_BORROW, /* class can't send but may borrow */ | |
85 | HTB_CAN_SEND /* class can send */ | |
1da177e4 LT |
86 | }; |
87 | ||
88 | /* interior & leaf nodes; props specific to leaves are marked L: */ | |
87990467 SH |
89 | struct htb_class { |
90 | /* general class parameters */ | |
91 | u32 classid; | |
92 | struct gnet_stats_basic bstats; | |
93 | struct gnet_stats_queue qstats; | |
94 | struct gnet_stats_rate_est rate_est; | |
95 | struct tc_htb_xstats xstats; /* our special stats */ | |
96 | int refcnt; /* usage count of this class */ | |
1da177e4 LT |
97 | |
98 | #ifdef HTB_RATECM | |
87990467 SH |
99 | /* rate measurement counters */ |
100 | unsigned long rate_bytes, sum_bytes; | |
101 | unsigned long rate_packets, sum_packets; | |
1da177e4 LT |
102 | #endif |
103 | ||
87990467 SH |
104 | /* topology */ |
105 | int level; /* our level (see above) */ | |
106 | struct htb_class *parent; /* parent class */ | |
0cef296d | 107 | struct hlist_node hlist; /* classid hash list item */ |
87990467 SH |
108 | struct list_head sibling; /* sibling list item */ |
109 | struct list_head children; /* children list */ | |
110 | ||
111 | union { | |
112 | struct htb_class_leaf { | |
113 | struct Qdisc *q; | |
114 | int prio; | |
115 | int aprio; | |
116 | int quantum; | |
117 | int deficit[TC_HTB_MAXDEPTH]; | |
118 | struct list_head drop_list; | |
119 | } leaf; | |
120 | struct htb_class_inner { | |
121 | struct rb_root feed[TC_HTB_NUMPRIO]; /* feed trees */ | |
122 | struct rb_node *ptr[TC_HTB_NUMPRIO]; /* current class ptr */ | |
123 | /* When class changes from state 1->2 and disconnects from | |
124 | parent's feed then we lost ptr value and start from the | |
125 | first child again. Here we store classid of the | |
126 | last valid ptr (used when ptr is NULL). */ | |
127 | u32 last_ptr_id[TC_HTB_NUMPRIO]; | |
128 | } inner; | |
129 | } un; | |
130 | struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ | |
131 | struct rb_node pq_node; /* node for event queue */ | |
fb983d45 | 132 | psched_time_t pq_key; |
87990467 SH |
133 | |
134 | int prio_activity; /* for which prios are we active */ | |
135 | enum htb_cmode cmode; /* current mode of the class */ | |
136 | ||
137 | /* class attached filters */ | |
138 | struct tcf_proto *filter_list; | |
139 | int filter_cnt; | |
140 | ||
141 | int warned; /* only one warning about non work conserving .. */ | |
142 | ||
143 | /* token bucket parameters */ | |
144 | struct qdisc_rate_table *rate; /* rate table of the class itself */ | |
145 | struct qdisc_rate_table *ceil; /* ceiling rate (limits borrows too) */ | |
146 | long buffer, cbuffer; /* token bucket depth/rate */ | |
147 | psched_tdiff_t mbuffer; /* max wait time */ | |
148 | long tokens, ctokens; /* current number of tokens */ | |
149 | psched_time_t t_c; /* checkpoint time */ | |
160d5e10 JP |
150 | |
151 | int prio; /* For parent to leaf return possible here */ | |
152 | int quantum; /* we do backup. Finally full replacement */ | |
153 | /* of un.leaf originals should be done. */ | |
1da177e4 LT |
154 | }; |
155 | ||
156 | /* TODO: maybe compute rate when size is too large .. or drop ? */ | |
87990467 SH |
157 | static inline long L2T(struct htb_class *cl, struct qdisc_rate_table *rate, |
158 | int size) | |
159 | { | |
160 | int slot = size >> rate->rate.cell_log; | |
161 | if (slot > 255) { | |
162 | cl->xstats.giants++; | |
163 | slot = 255; | |
164 | } | |
165 | return rate->data[slot]; | |
1da177e4 LT |
166 | } |
167 | ||
87990467 SH |
168 | struct htb_sched { |
169 | struct list_head root; /* root classes list */ | |
0cef296d SH |
170 | struct hlist_head hash[HTB_HSIZE]; /* hashed by classid */ |
171 | struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */ | |
87990467 SH |
172 | |
173 | /* self list - roots of self generating tree */ | |
174 | struct rb_root row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
175 | int row_mask[TC_HTB_MAXDEPTH]; | |
176 | struct rb_node *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
177 | u32 last_ptr_id[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; | |
1da177e4 | 178 | |
87990467 SH |
179 | /* self wait list - roots of wait PQs per row */ |
180 | struct rb_root wait_pq[TC_HTB_MAXDEPTH]; | |
1da177e4 | 181 | |
87990467 | 182 | /* time of nearest event per level (row) */ |
fb983d45 | 183 | psched_time_t near_ev_cache[TC_HTB_MAXDEPTH]; |
1da177e4 | 184 | |
87990467 SH |
185 | /* whether we hit non-work conserving class during this dequeue; we use */ |
186 | int nwc_hit; /* this to disable mindelay complaint in dequeue */ | |
1da177e4 | 187 | |
87990467 | 188 | int defcls; /* class where unclassified flows go to */ |
1da177e4 | 189 | |
87990467 SH |
190 | /* filters for qdisc itself */ |
191 | struct tcf_proto *filter_list; | |
192 | int filter_cnt; | |
1da177e4 | 193 | |
87990467 SH |
194 | int rate2quantum; /* quant = rate / rate2quantum */ |
195 | psched_time_t now; /* cached dequeue time */ | |
fb983d45 | 196 | struct qdisc_watchdog watchdog; |
1da177e4 | 197 | #ifdef HTB_RATECM |
87990467 SH |
198 | struct timer_list rttim; /* rate computer timer */ |
199 | int recmp_bucket; /* which hash bucket to recompute next */ | |
1da177e4 | 200 | #endif |
1da177e4 | 201 | |
87990467 SH |
202 | /* non shaped skbs; let them go directly thru */ |
203 | struct sk_buff_head direct_queue; | |
204 | int direct_qlen; /* max qlen of above */ | |
205 | ||
206 | long direct_pkts; | |
1da177e4 LT |
207 | }; |
208 | ||
209 | /* compute hash of size HTB_HSIZE for given handle */ | |
87990467 | 210 | static inline int htb_hash(u32 h) |
1da177e4 LT |
211 | { |
212 | #if HTB_HSIZE != 16 | |
87990467 | 213 | #error "Declare new hash for your HTB_HSIZE" |
1da177e4 | 214 | #endif |
87990467 SH |
215 | h ^= h >> 8; /* stolen from cbq_hash */ |
216 | h ^= h >> 4; | |
217 | return h & 0xf; | |
1da177e4 LT |
218 | } |
219 | ||
220 | /* find class in global hash table using given handle */ | |
87990467 | 221 | static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch) |
1da177e4 LT |
222 | { |
223 | struct htb_sched *q = qdisc_priv(sch); | |
0cef296d SH |
224 | struct hlist_node *p; |
225 | struct htb_class *cl; | |
226 | ||
87990467 | 227 | if (TC_H_MAJ(handle) != sch->handle) |
1da177e4 | 228 | return NULL; |
87990467 | 229 | |
0cef296d | 230 | hlist_for_each_entry(cl, p, q->hash + htb_hash(handle), hlist) { |
1da177e4 LT |
231 | if (cl->classid == handle) |
232 | return cl; | |
233 | } | |
234 | return NULL; | |
235 | } | |
236 | ||
237 | /** | |
238 | * htb_classify - classify a packet into class | |
239 | * | |
240 | * It returns NULL if the packet should be dropped or -1 if the packet | |
241 | * should be passed directly thru. In all other cases leaf class is returned. | |
242 | * We allow direct class selection by classid in priority. The we examine | |
243 | * filters in qdisc and in inner nodes (if higher filter points to the inner | |
244 | * node). If we end up with classid MAJOR:0 we enqueue the skb into special | |
10297b99 | 245 | * internal fifo (direct). These packets then go directly thru. If we still |
1da177e4 LT |
246 | * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessfull |
247 | * then finish and return direct queue. | |
248 | */ | |
249 | #define HTB_DIRECT (struct htb_class*)-1 | |
250 | static inline u32 htb_classid(struct htb_class *cl) | |
251 | { | |
252 | return (cl && cl != HTB_DIRECT) ? cl->classid : TC_H_UNSPEC; | |
253 | } | |
254 | ||
87990467 SH |
255 | static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, |
256 | int *qerr) | |
1da177e4 LT |
257 | { |
258 | struct htb_sched *q = qdisc_priv(sch); | |
259 | struct htb_class *cl; | |
260 | struct tcf_result res; | |
261 | struct tcf_proto *tcf; | |
262 | int result; | |
263 | ||
264 | /* allow to select class by setting skb->priority to valid classid; | |
265 | note that nfmark can be used too by attaching filter fw with no | |
266 | rules in it */ | |
267 | if (skb->priority == sch->handle) | |
87990467 SH |
268 | return HTB_DIRECT; /* X:0 (direct flow) selected */ |
269 | if ((cl = htb_find(skb->priority, sch)) != NULL && cl->level == 0) | |
1da177e4 LT |
270 | return cl; |
271 | ||
29f1df6c | 272 | *qerr = NET_XMIT_BYPASS; |
1da177e4 LT |
273 | tcf = q->filter_list; |
274 | while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) { | |
275 | #ifdef CONFIG_NET_CLS_ACT | |
276 | switch (result) { | |
277 | case TC_ACT_QUEUED: | |
87990467 | 278 | case TC_ACT_STOLEN: |
1da177e4 LT |
279 | *qerr = NET_XMIT_SUCCESS; |
280 | case TC_ACT_SHOT: | |
281 | return NULL; | |
282 | } | |
283 | #elif defined(CONFIG_NET_CLS_POLICE) | |
284 | if (result == TC_POLICE_SHOT) | |
285 | return HTB_DIRECT; | |
286 | #endif | |
87990467 | 287 | if ((cl = (void *)res.class) == NULL) { |
1da177e4 | 288 | if (res.classid == sch->handle) |
87990467 SH |
289 | return HTB_DIRECT; /* X:0 (direct flow) */ |
290 | if ((cl = htb_find(res.classid, sch)) == NULL) | |
291 | break; /* filter selected invalid classid */ | |
1da177e4 LT |
292 | } |
293 | if (!cl->level) | |
87990467 | 294 | return cl; /* we hit leaf; return it */ |
1da177e4 LT |
295 | |
296 | /* we have got inner class; apply inner filter chain */ | |
297 | tcf = cl->filter_list; | |
298 | } | |
299 | /* classification failed; try to use default class */ | |
87990467 | 300 | cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch); |
1da177e4 | 301 | if (!cl || cl->level) |
87990467 | 302 | return HTB_DIRECT; /* bad default .. this is safe bet */ |
1da177e4 LT |
303 | return cl; |
304 | } | |
305 | ||
1da177e4 LT |
306 | /** |
307 | * htb_add_to_id_tree - adds class to the round robin list | |
308 | * | |
309 | * Routine adds class to the list (actually tree) sorted by classid. | |
310 | * Make sure that class is not already on such list for given prio. | |
311 | */ | |
87990467 SH |
312 | static void htb_add_to_id_tree(struct rb_root *root, |
313 | struct htb_class *cl, int prio) | |
1da177e4 LT |
314 | { |
315 | struct rb_node **p = &root->rb_node, *parent = NULL; | |
3bf72957 | 316 | |
1da177e4 | 317 | while (*p) { |
87990467 SH |
318 | struct htb_class *c; |
319 | parent = *p; | |
1da177e4 | 320 | c = rb_entry(parent, struct htb_class, node[prio]); |
3bf72957 | 321 | |
1da177e4 LT |
322 | if (cl->classid > c->classid) |
323 | p = &parent->rb_right; | |
87990467 | 324 | else |
1da177e4 LT |
325 | p = &parent->rb_left; |
326 | } | |
327 | rb_link_node(&cl->node[prio], parent, p); | |
328 | rb_insert_color(&cl->node[prio], root); | |
329 | } | |
330 | ||
331 | /** | |
332 | * htb_add_to_wait_tree - adds class to the event queue with delay | |
333 | * | |
334 | * The class is added to priority event queue to indicate that class will | |
335 | * change its mode in cl->pq_key microseconds. Make sure that class is not | |
336 | * already in the queue. | |
337 | */ | |
87990467 SH |
338 | static void htb_add_to_wait_tree(struct htb_sched *q, |
339 | struct htb_class *cl, long delay) | |
1da177e4 LT |
340 | { |
341 | struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL; | |
3bf72957 | 342 | |
fb983d45 PM |
343 | cl->pq_key = q->now + delay; |
344 | if (cl->pq_key == q->now) | |
1da177e4 LT |
345 | cl->pq_key++; |
346 | ||
347 | /* update the nearest event cache */ | |
fb983d45 | 348 | if (q->near_ev_cache[cl->level] > cl->pq_key) |
1da177e4 | 349 | q->near_ev_cache[cl->level] = cl->pq_key; |
87990467 | 350 | |
1da177e4 | 351 | while (*p) { |
87990467 SH |
352 | struct htb_class *c; |
353 | parent = *p; | |
1da177e4 | 354 | c = rb_entry(parent, struct htb_class, pq_node); |
fb983d45 | 355 | if (cl->pq_key >= c->pq_key) |
1da177e4 | 356 | p = &parent->rb_right; |
87990467 | 357 | else |
1da177e4 LT |
358 | p = &parent->rb_left; |
359 | } | |
360 | rb_link_node(&cl->pq_node, parent, p); | |
361 | rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]); | |
362 | } | |
363 | ||
364 | /** | |
365 | * htb_next_rb_node - finds next node in binary tree | |
366 | * | |
367 | * When we are past last key we return NULL. | |
368 | * Average complexity is 2 steps per call. | |
369 | */ | |
3696f625 | 370 | static inline void htb_next_rb_node(struct rb_node **n) |
1da177e4 LT |
371 | { |
372 | *n = rb_next(*n); | |
373 | } | |
374 | ||
375 | /** | |
376 | * htb_add_class_to_row - add class to its row | |
377 | * | |
378 | * The class is added to row at priorities marked in mask. | |
379 | * It does nothing if mask == 0. | |
380 | */ | |
87990467 SH |
381 | static inline void htb_add_class_to_row(struct htb_sched *q, |
382 | struct htb_class *cl, int mask) | |
1da177e4 | 383 | { |
1da177e4 LT |
384 | q->row_mask[cl->level] |= mask; |
385 | while (mask) { | |
386 | int prio = ffz(~mask); | |
387 | mask &= ~(1 << prio); | |
87990467 | 388 | htb_add_to_id_tree(q->row[cl->level] + prio, cl, prio); |
1da177e4 LT |
389 | } |
390 | } | |
391 | ||
3696f625 SH |
392 | /* If this triggers, it is a bug in this code, but it need not be fatal */ |
393 | static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root) | |
394 | { | |
81771b3b | 395 | if (RB_EMPTY_NODE(rb)) { |
3696f625 SH |
396 | WARN_ON(1); |
397 | } else { | |
398 | rb_erase(rb, root); | |
399 | RB_CLEAR_NODE(rb); | |
400 | } | |
401 | } | |
402 | ||
403 | ||
1da177e4 LT |
404 | /** |
405 | * htb_remove_class_from_row - removes class from its row | |
406 | * | |
407 | * The class is removed from row at priorities marked in mask. | |
408 | * It does nothing if mask == 0. | |
409 | */ | |
87990467 SH |
410 | static inline void htb_remove_class_from_row(struct htb_sched *q, |
411 | struct htb_class *cl, int mask) | |
1da177e4 LT |
412 | { |
413 | int m = 0; | |
3bf72957 | 414 | |
1da177e4 LT |
415 | while (mask) { |
416 | int prio = ffz(~mask); | |
3696f625 | 417 | |
1da177e4 | 418 | mask &= ~(1 << prio); |
87990467 SH |
419 | if (q->ptr[cl->level][prio] == cl->node + prio) |
420 | htb_next_rb_node(q->ptr[cl->level] + prio); | |
3696f625 SH |
421 | |
422 | htb_safe_rb_erase(cl->node + prio, q->row[cl->level] + prio); | |
87990467 | 423 | if (!q->row[cl->level][prio].rb_node) |
1da177e4 LT |
424 | m |= 1 << prio; |
425 | } | |
1da177e4 LT |
426 | q->row_mask[cl->level] &= ~m; |
427 | } | |
428 | ||
429 | /** | |
430 | * htb_activate_prios - creates active classe's feed chain | |
431 | * | |
432 | * The class is connected to ancestors and/or appropriate rows | |
10297b99 | 433 | * for priorities it is participating on. cl->cmode must be new |
1da177e4 LT |
434 | * (activated) mode. It does nothing if cl->prio_activity == 0. |
435 | */ | |
87990467 | 436 | static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
437 | { |
438 | struct htb_class *p = cl->parent; | |
87990467 | 439 | long m, mask = cl->prio_activity; |
1da177e4 LT |
440 | |
441 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
442 | m = mask; |
443 | while (m) { | |
1da177e4 LT |
444 | int prio = ffz(~m); |
445 | m &= ~(1 << prio); | |
87990467 | 446 | |
1da177e4 LT |
447 | if (p->un.inner.feed[prio].rb_node) |
448 | /* parent already has its feed in use so that | |
449 | reset bit in mask as parent is already ok */ | |
450 | mask &= ~(1 << prio); | |
87990467 SH |
451 | |
452 | htb_add_to_id_tree(p->un.inner.feed + prio, cl, prio); | |
1da177e4 | 453 | } |
1da177e4 | 454 | p->prio_activity |= mask; |
87990467 SH |
455 | cl = p; |
456 | p = cl->parent; | |
3bf72957 | 457 | |
1da177e4 LT |
458 | } |
459 | if (cl->cmode == HTB_CAN_SEND && mask) | |
87990467 | 460 | htb_add_class_to_row(q, cl, mask); |
1da177e4 LT |
461 | } |
462 | ||
463 | /** | |
464 | * htb_deactivate_prios - remove class from feed chain | |
465 | * | |
10297b99 | 466 | * cl->cmode must represent old mode (before deactivation). It does |
1da177e4 LT |
467 | * nothing if cl->prio_activity == 0. Class is removed from all feed |
468 | * chains and rows. | |
469 | */ | |
470 | static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) | |
471 | { | |
472 | struct htb_class *p = cl->parent; | |
87990467 | 473 | long m, mask = cl->prio_activity; |
1da177e4 LT |
474 | |
475 | while (cl->cmode == HTB_MAY_BORROW && p && mask) { | |
87990467 SH |
476 | m = mask; |
477 | mask = 0; | |
1da177e4 LT |
478 | while (m) { |
479 | int prio = ffz(~m); | |
480 | m &= ~(1 << prio); | |
87990467 SH |
481 | |
482 | if (p->un.inner.ptr[prio] == cl->node + prio) { | |
1da177e4 LT |
483 | /* we are removing child which is pointed to from |
484 | parent feed - forget the pointer but remember | |
485 | classid */ | |
486 | p->un.inner.last_ptr_id[prio] = cl->classid; | |
487 | p->un.inner.ptr[prio] = NULL; | |
488 | } | |
87990467 | 489 | |
3696f625 | 490 | htb_safe_rb_erase(cl->node + prio, p->un.inner.feed + prio); |
87990467 SH |
491 | |
492 | if (!p->un.inner.feed[prio].rb_node) | |
1da177e4 LT |
493 | mask |= 1 << prio; |
494 | } | |
3bf72957 | 495 | |
1da177e4 | 496 | p->prio_activity &= ~mask; |
87990467 SH |
497 | cl = p; |
498 | p = cl->parent; | |
3bf72957 | 499 | |
1da177e4 | 500 | } |
87990467 SH |
501 | if (cl->cmode == HTB_CAN_SEND && mask) |
502 | htb_remove_class_from_row(q, cl, mask); | |
1da177e4 LT |
503 | } |
504 | ||
18a63e86 SH |
505 | #if HTB_HYSTERESIS |
506 | static inline long htb_lowater(const struct htb_class *cl) | |
507 | { | |
508 | return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0; | |
509 | } | |
510 | static inline long htb_hiwater(const struct htb_class *cl) | |
511 | { | |
512 | return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0; | |
513 | } | |
514 | #else | |
515 | #define htb_lowater(cl) (0) | |
516 | #define htb_hiwater(cl) (0) | |
517 | #endif | |
518 | ||
1da177e4 LT |
519 | /** |
520 | * htb_class_mode - computes and returns current class mode | |
521 | * | |
522 | * It computes cl's mode at time cl->t_c+diff and returns it. If mode | |
523 | * is not HTB_CAN_SEND then cl->pq_key is updated to time difference | |
10297b99 | 524 | * from now to time when cl will change its state. |
1da177e4 | 525 | * Also it is worth to note that class mode doesn't change simply |
10297b99 | 526 | * at cl->{c,}tokens == 0 but there can rather be hysteresis of |
1da177e4 LT |
527 | * 0 .. -cl->{c,}buffer range. It is meant to limit number of |
528 | * mode transitions per time unit. The speed gain is about 1/6. | |
529 | */ | |
87990467 SH |
530 | static inline enum htb_cmode |
531 | htb_class_mode(struct htb_class *cl, long *diff) | |
1da177e4 | 532 | { |
87990467 | 533 | long toks; |
1da177e4 | 534 | |
87990467 SH |
535 | if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) { |
536 | *diff = -toks; | |
537 | return HTB_CANT_SEND; | |
538 | } | |
18a63e86 | 539 | |
87990467 SH |
540 | if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl)) |
541 | return HTB_CAN_SEND; | |
1da177e4 | 542 | |
87990467 SH |
543 | *diff = -toks; |
544 | return HTB_MAY_BORROW; | |
1da177e4 LT |
545 | } |
546 | ||
547 | /** | |
548 | * htb_change_class_mode - changes classe's mode | |
549 | * | |
550 | * This should be the only way how to change classe's mode under normal | |
551 | * cirsumstances. Routine will update feed lists linkage, change mode | |
552 | * and add class to the wait event queue if appropriate. New mode should | |
553 | * be different from old one and cl->pq_key has to be valid if changing | |
554 | * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). | |
555 | */ | |
87990467 | 556 | static void |
1da177e4 | 557 | htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff) |
87990467 SH |
558 | { |
559 | enum htb_cmode new_mode = htb_class_mode(cl, diff); | |
1da177e4 LT |
560 | |
561 | if (new_mode == cl->cmode) | |
87990467 SH |
562 | return; |
563 | ||
564 | if (cl->prio_activity) { /* not necessary: speed optimization */ | |
565 | if (cl->cmode != HTB_CANT_SEND) | |
566 | htb_deactivate_prios(q, cl); | |
1da177e4 | 567 | cl->cmode = new_mode; |
87990467 SH |
568 | if (new_mode != HTB_CANT_SEND) |
569 | htb_activate_prios(q, cl); | |
570 | } else | |
1da177e4 LT |
571 | cl->cmode = new_mode; |
572 | } | |
573 | ||
574 | /** | |
10297b99 | 575 | * htb_activate - inserts leaf cl into appropriate active feeds |
1da177e4 LT |
576 | * |
577 | * Routine learns (new) priority of leaf and activates feed chain | |
578 | * for the prio. It can be called on already active leaf safely. | |
579 | * It also adds leaf into droplist. | |
580 | */ | |
87990467 | 581 | static inline void htb_activate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
582 | { |
583 | BUG_TRAP(!cl->level && cl->un.leaf.q && cl->un.leaf.q->q.qlen); | |
3bf72957 | 584 | |
1da177e4 LT |
585 | if (!cl->prio_activity) { |
586 | cl->prio_activity = 1 << (cl->un.leaf.aprio = cl->un.leaf.prio); | |
87990467 SH |
587 | htb_activate_prios(q, cl); |
588 | list_add_tail(&cl->un.leaf.drop_list, | |
589 | q->drops + cl->un.leaf.aprio); | |
1da177e4 LT |
590 | } |
591 | } | |
592 | ||
593 | /** | |
10297b99 | 594 | * htb_deactivate - remove leaf cl from active feeds |
1da177e4 LT |
595 | * |
596 | * Make sure that leaf is active. In the other words it can't be called | |
597 | * with non-active leaf. It also removes class from the drop list. | |
598 | */ | |
87990467 | 599 | static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl) |
1da177e4 LT |
600 | { |
601 | BUG_TRAP(cl->prio_activity); | |
3bf72957 | 602 | |
87990467 | 603 | htb_deactivate_prios(q, cl); |
1da177e4 LT |
604 | cl->prio_activity = 0; |
605 | list_del_init(&cl->un.leaf.drop_list); | |
606 | } | |
607 | ||
608 | static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch) | |
609 | { | |
87990467 SH |
610 | int ret; |
611 | struct htb_sched *q = qdisc_priv(sch); | |
612 | struct htb_class *cl = htb_classify(skb, sch, &ret); | |
613 | ||
614 | if (cl == HTB_DIRECT) { | |
615 | /* enqueue to helper queue */ | |
616 | if (q->direct_queue.qlen < q->direct_qlen) { | |
617 | __skb_queue_tail(&q->direct_queue, skb); | |
618 | q->direct_pkts++; | |
619 | } else { | |
620 | kfree_skb(skb); | |
621 | sch->qstats.drops++; | |
622 | return NET_XMIT_DROP; | |
623 | } | |
1da177e4 | 624 | #ifdef CONFIG_NET_CLS_ACT |
87990467 SH |
625 | } else if (!cl) { |
626 | if (ret == NET_XMIT_BYPASS) | |
627 | sch->qstats.drops++; | |
628 | kfree_skb(skb); | |
629 | return ret; | |
1da177e4 | 630 | #endif |
87990467 SH |
631 | } else if (cl->un.leaf.q->enqueue(skb, cl->un.leaf.q) != |
632 | NET_XMIT_SUCCESS) { | |
633 | sch->qstats.drops++; | |
634 | cl->qstats.drops++; | |
635 | return NET_XMIT_DROP; | |
636 | } else { | |
637 | cl->bstats.packets++; | |
638 | cl->bstats.bytes += skb->len; | |
639 | htb_activate(q, cl); | |
640 | } | |
641 | ||
642 | sch->q.qlen++; | |
643 | sch->bstats.packets++; | |
644 | sch->bstats.bytes += skb->len; | |
645 | return NET_XMIT_SUCCESS; | |
1da177e4 LT |
646 | } |
647 | ||
648 | /* TODO: requeuing packet charges it to policers again !! */ | |
649 | static int htb_requeue(struct sk_buff *skb, struct Qdisc *sch) | |
650 | { | |
87990467 SH |
651 | struct htb_sched *q = qdisc_priv(sch); |
652 | int ret = NET_XMIT_SUCCESS; | |
653 | struct htb_class *cl = htb_classify(skb, sch, &ret); | |
654 | struct sk_buff *tskb; | |
655 | ||
656 | if (cl == HTB_DIRECT || !cl) { | |
657 | /* enqueue to helper queue */ | |
658 | if (q->direct_queue.qlen < q->direct_qlen && cl) { | |
659 | __skb_queue_head(&q->direct_queue, skb); | |
660 | } else { | |
661 | __skb_queue_head(&q->direct_queue, skb); | |
662 | tskb = __skb_dequeue_tail(&q->direct_queue); | |
663 | kfree_skb(tskb); | |
664 | sch->qstats.drops++; | |
665 | return NET_XMIT_CN; | |
666 | } | |
667 | } else if (cl->un.leaf.q->ops->requeue(skb, cl->un.leaf.q) != | |
668 | NET_XMIT_SUCCESS) { | |
669 | sch->qstats.drops++; | |
670 | cl->qstats.drops++; | |
671 | return NET_XMIT_DROP; | |
672 | } else | |
673 | htb_activate(q, cl); | |
674 | ||
675 | sch->q.qlen++; | |
676 | sch->qstats.requeues++; | |
677 | return NET_XMIT_SUCCESS; | |
1da177e4 LT |
678 | } |
679 | ||
1da177e4 LT |
680 | #ifdef HTB_RATECM |
681 | #define RT_GEN(D,R) R+=D-(R/HTB_EWMAC);D=0 | |
682 | static void htb_rate_timer(unsigned long arg) | |
683 | { | |
87990467 | 684 | struct Qdisc *sch = (struct Qdisc *)arg; |
1da177e4 | 685 | struct htb_sched *q = qdisc_priv(sch); |
0cef296d SH |
686 | struct hlist_node *p; |
687 | struct htb_class *cl; | |
688 | ||
1da177e4 LT |
689 | |
690 | /* lock queue so that we can muck with it */ | |
9ac961ee | 691 | spin_lock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
692 | |
693 | q->rttim.expires = jiffies + HZ; | |
694 | add_timer(&q->rttim); | |
695 | ||
696 | /* scan and recompute one bucket at time */ | |
87990467 | 697 | if (++q->recmp_bucket >= HTB_HSIZE) |
1da177e4 | 698 | q->recmp_bucket = 0; |
3bf72957 | 699 | |
0cef296d | 700 | hlist_for_each_entry(cl,p, q->hash + q->recmp_bucket, hlist) { |
87990467 SH |
701 | RT_GEN(cl->sum_bytes, cl->rate_bytes); |
702 | RT_GEN(cl->sum_packets, cl->rate_packets); | |
1da177e4 | 703 | } |
9ac961ee | 704 | spin_unlock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
705 | } |
706 | #endif | |
707 | ||
708 | /** | |
709 | * htb_charge_class - charges amount "bytes" to leaf and ancestors | |
710 | * | |
711 | * Routine assumes that packet "bytes" long was dequeued from leaf cl | |
712 | * borrowing from "level". It accounts bytes to ceil leaky bucket for | |
713 | * leaf and all ancestors and to rate bucket for ancestors at levels | |
714 | * "level" and higher. It also handles possible change of mode resulting | |
715 | * from the update. Note that mode can also increase here (MAY_BORROW to | |
716 | * CAN_SEND) because we can use more precise clock that event queue here. | |
717 | * In such case we remove class from event queue first. | |
718 | */ | |
87990467 SH |
719 | static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, |
720 | int level, int bytes) | |
721 | { | |
722 | long toks, diff; | |
1da177e4 | 723 | enum htb_cmode old_mode; |
1da177e4 LT |
724 | |
725 | #define HTB_ACCNT(T,B,R) toks = diff + cl->T; \ | |
726 | if (toks > cl->B) toks = cl->B; \ | |
727 | toks -= L2T(cl, cl->R, bytes); \ | |
728 | if (toks <= -cl->mbuffer) toks = 1-cl->mbuffer; \ | |
729 | cl->T = toks | |
730 | ||
731 | while (cl) { | |
87990467 | 732 | diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32) cl->mbuffer); |
1da177e4 | 733 | if (cl->level >= level) { |
87990467 SH |
734 | if (cl->level == level) |
735 | cl->xstats.lends++; | |
736 | HTB_ACCNT(tokens, buffer, rate); | |
1da177e4 LT |
737 | } else { |
738 | cl->xstats.borrows++; | |
87990467 | 739 | cl->tokens += diff; /* we moved t_c; update tokens */ |
1da177e4 | 740 | } |
87990467 | 741 | HTB_ACCNT(ctokens, cbuffer, ceil); |
1da177e4 | 742 | cl->t_c = q->now; |
1da177e4 | 743 | |
87990467 SH |
744 | old_mode = cl->cmode; |
745 | diff = 0; | |
746 | htb_change_class_mode(q, cl, &diff); | |
1da177e4 LT |
747 | if (old_mode != cl->cmode) { |
748 | if (old_mode != HTB_CAN_SEND) | |
3696f625 | 749 | htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level); |
1da177e4 | 750 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 751 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 | 752 | } |
1da177e4 LT |
753 | #ifdef HTB_RATECM |
754 | /* update rate counters */ | |
87990467 SH |
755 | cl->sum_bytes += bytes; |
756 | cl->sum_packets++; | |
1da177e4 LT |
757 | #endif |
758 | ||
759 | /* update byte stats except for leaves which are already updated */ | |
760 | if (cl->level) { | |
761 | cl->bstats.bytes += bytes; | |
762 | cl->bstats.packets++; | |
763 | } | |
764 | cl = cl->parent; | |
765 | } | |
766 | } | |
767 | ||
768 | /** | |
769 | * htb_do_events - make mode changes to classes at the level | |
770 | * | |
fb983d45 | 771 | * Scans event queue for pending events and applies them. Returns time of |
1da177e4 | 772 | * next pending event (0 for no event in pq). |
fb983d45 | 773 | * Note: Applied are events whose have cl->pq_key <= q->now. |
1da177e4 | 774 | */ |
fb983d45 | 775 | static psched_time_t htb_do_events(struct htb_sched *q, int level) |
1da177e4 LT |
776 | { |
777 | int i; | |
3bf72957 | 778 | |
1da177e4 LT |
779 | for (i = 0; i < 500; i++) { |
780 | struct htb_class *cl; | |
781 | long diff; | |
30bdbe39 AM |
782 | struct rb_node *p = rb_first(&q->wait_pq[level]); |
783 | ||
87990467 SH |
784 | if (!p) |
785 | return 0; | |
1da177e4 LT |
786 | |
787 | cl = rb_entry(p, struct htb_class, pq_node); | |
fb983d45 PM |
788 | if (cl->pq_key > q->now) |
789 | return cl->pq_key; | |
790 | ||
3696f625 | 791 | htb_safe_rb_erase(p, q->wait_pq + level); |
87990467 SH |
792 | diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32) cl->mbuffer); |
793 | htb_change_class_mode(q, cl, &diff); | |
1da177e4 | 794 | if (cl->cmode != HTB_CAN_SEND) |
87990467 | 795 | htb_add_to_wait_tree(q, cl, diff); |
1da177e4 LT |
796 | } |
797 | if (net_ratelimit()) | |
798 | printk(KERN_WARNING "htb: too many events !\n"); | |
fb983d45 | 799 | return q->now + PSCHED_TICKS_PER_SEC / 10; |
1da177e4 LT |
800 | } |
801 | ||
802 | /* Returns class->node+prio from id-tree where classe's id is >= id. NULL | |
803 | is no such one exists. */ | |
87990467 SH |
804 | static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n, |
805 | u32 id) | |
1da177e4 LT |
806 | { |
807 | struct rb_node *r = NULL; | |
808 | while (n) { | |
87990467 SH |
809 | struct htb_class *cl = |
810 | rb_entry(n, struct htb_class, node[prio]); | |
811 | if (id == cl->classid) | |
812 | return n; | |
813 | ||
1da177e4 LT |
814 | if (id > cl->classid) { |
815 | n = n->rb_right; | |
816 | } else { | |
817 | r = n; | |
818 | n = n->rb_left; | |
819 | } | |
820 | } | |
821 | return r; | |
822 | } | |
823 | ||
824 | /** | |
825 | * htb_lookup_leaf - returns next leaf class in DRR order | |
826 | * | |
827 | * Find leaf where current feed pointers points to. | |
828 | */ | |
87990467 SH |
829 | static struct htb_class *htb_lookup_leaf(struct rb_root *tree, int prio, |
830 | struct rb_node **pptr, u32 * pid) | |
1da177e4 LT |
831 | { |
832 | int i; | |
833 | struct { | |
834 | struct rb_node *root; | |
835 | struct rb_node **pptr; | |
836 | u32 *pid; | |
87990467 SH |
837 | } stk[TC_HTB_MAXDEPTH], *sp = stk; |
838 | ||
1da177e4 LT |
839 | BUG_TRAP(tree->rb_node); |
840 | sp->root = tree->rb_node; | |
841 | sp->pptr = pptr; | |
842 | sp->pid = pid; | |
843 | ||
844 | for (i = 0; i < 65535; i++) { | |
87990467 | 845 | if (!*sp->pptr && *sp->pid) { |
10297b99 | 846 | /* ptr was invalidated but id is valid - try to recover |
1da177e4 | 847 | the original or next ptr */ |
87990467 SH |
848 | *sp->pptr = |
849 | htb_id_find_next_upper(prio, sp->root, *sp->pid); | |
1da177e4 | 850 | } |
87990467 SH |
851 | *sp->pid = 0; /* ptr is valid now so that remove this hint as it |
852 | can become out of date quickly */ | |
853 | if (!*sp->pptr) { /* we are at right end; rewind & go up */ | |
1da177e4 | 854 | *sp->pptr = sp->root; |
87990467 | 855 | while ((*sp->pptr)->rb_left) |
1da177e4 LT |
856 | *sp->pptr = (*sp->pptr)->rb_left; |
857 | if (sp > stk) { | |
858 | sp--; | |
87990467 SH |
859 | BUG_TRAP(*sp->pptr); |
860 | if (!*sp->pptr) | |
861 | return NULL; | |
862 | htb_next_rb_node(sp->pptr); | |
1da177e4 LT |
863 | } |
864 | } else { | |
865 | struct htb_class *cl; | |
87990467 SH |
866 | cl = rb_entry(*sp->pptr, struct htb_class, node[prio]); |
867 | if (!cl->level) | |
1da177e4 LT |
868 | return cl; |
869 | (++sp)->root = cl->un.inner.feed[prio].rb_node; | |
87990467 SH |
870 | sp->pptr = cl->un.inner.ptr + prio; |
871 | sp->pid = cl->un.inner.last_ptr_id + prio; | |
1da177e4 LT |
872 | } |
873 | } | |
874 | BUG_TRAP(0); | |
875 | return NULL; | |
876 | } | |
877 | ||
878 | /* dequeues packet at given priority and level; call only if | |
879 | you are sure that there is active class at prio/level */ | |
87990467 SH |
880 | static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, int prio, |
881 | int level) | |
1da177e4 LT |
882 | { |
883 | struct sk_buff *skb = NULL; | |
87990467 | 884 | struct htb_class *cl, *start; |
1da177e4 | 885 | /* look initial class up in the row */ |
87990467 SH |
886 | start = cl = htb_lookup_leaf(q->row[level] + prio, prio, |
887 | q->ptr[level] + prio, | |
888 | q->last_ptr_id[level] + prio); | |
889 | ||
1da177e4 LT |
890 | do { |
891 | next: | |
87990467 SH |
892 | BUG_TRAP(cl); |
893 | if (!cl) | |
894 | return NULL; | |
1da177e4 LT |
895 | |
896 | /* class can be empty - it is unlikely but can be true if leaf | |
897 | qdisc drops packets in enqueue routine or if someone used | |
10297b99 | 898 | graft operation on the leaf since last dequeue; |
1da177e4 LT |
899 | simply deactivate and skip such class */ |
900 | if (unlikely(cl->un.leaf.q->q.qlen == 0)) { | |
901 | struct htb_class *next; | |
87990467 | 902 | htb_deactivate(q, cl); |
1da177e4 LT |
903 | |
904 | /* row/level might become empty */ | |
905 | if ((q->row_mask[level] & (1 << prio)) == 0) | |
87990467 | 906 | return NULL; |
1da177e4 | 907 | |
87990467 SH |
908 | next = htb_lookup_leaf(q->row[level] + prio, |
909 | prio, q->ptr[level] + prio, | |
910 | q->last_ptr_id[level] + prio); | |
911 | ||
912 | if (cl == start) /* fix start if we just deleted it */ | |
1da177e4 LT |
913 | start = next; |
914 | cl = next; | |
915 | goto next; | |
916 | } | |
87990467 SH |
917 | |
918 | skb = cl->un.leaf.q->dequeue(cl->un.leaf.q); | |
919 | if (likely(skb != NULL)) | |
1da177e4 LT |
920 | break; |
921 | if (!cl->warned) { | |
87990467 SH |
922 | printk(KERN_WARNING |
923 | "htb: class %X isn't work conserving ?!\n", | |
924 | cl->classid); | |
1da177e4 LT |
925 | cl->warned = 1; |
926 | } | |
927 | q->nwc_hit++; | |
87990467 SH |
928 | htb_next_rb_node((level ? cl->parent->un.inner.ptr : q-> |
929 | ptr[0]) + prio); | |
930 | cl = htb_lookup_leaf(q->row[level] + prio, prio, | |
931 | q->ptr[level] + prio, | |
932 | q->last_ptr_id[level] + prio); | |
1da177e4 LT |
933 | |
934 | } while (cl != start); | |
935 | ||
936 | if (likely(skb != NULL)) { | |
937 | if ((cl->un.leaf.deficit[level] -= skb->len) < 0) { | |
1da177e4 | 938 | cl->un.leaf.deficit[level] += cl->un.leaf.quantum; |
87990467 SH |
939 | htb_next_rb_node((level ? cl->parent->un.inner.ptr : q-> |
940 | ptr[0]) + prio); | |
1da177e4 LT |
941 | } |
942 | /* this used to be after charge_class but this constelation | |
943 | gives us slightly better performance */ | |
944 | if (!cl->un.leaf.q->q.qlen) | |
87990467 SH |
945 | htb_deactivate(q, cl); |
946 | htb_charge_class(q, cl, level, skb->len); | |
1da177e4 LT |
947 | } |
948 | return skb; | |
949 | } | |
950 | ||
1da177e4 LT |
951 | static struct sk_buff *htb_dequeue(struct Qdisc *sch) |
952 | { | |
953 | struct sk_buff *skb = NULL; | |
954 | struct htb_sched *q = qdisc_priv(sch); | |
955 | int level; | |
fb983d45 | 956 | psched_time_t next_event; |
1da177e4 LT |
957 | |
958 | /* try to dequeue direct packets as high prio (!) to minimize cpu work */ | |
87990467 SH |
959 | skb = __skb_dequeue(&q->direct_queue); |
960 | if (skb != NULL) { | |
1da177e4 LT |
961 | sch->flags &= ~TCQ_F_THROTTLED; |
962 | sch->q.qlen--; | |
963 | return skb; | |
964 | } | |
965 | ||
87990467 SH |
966 | if (!sch->q.qlen) |
967 | goto fin; | |
1da177e4 LT |
968 | PSCHED_GET_TIME(q->now); |
969 | ||
fb983d45 | 970 | next_event = q->now + 5 * PSCHED_TICKS_PER_SEC; |
1da177e4 LT |
971 | q->nwc_hit = 0; |
972 | for (level = 0; level < TC_HTB_MAXDEPTH; level++) { | |
973 | /* common case optimization - skip event handler quickly */ | |
974 | int m; | |
fb983d45 PM |
975 | psched_time_t event; |
976 | ||
977 | if (q->now >= q->near_ev_cache[level]) { | |
978 | event = htb_do_events(q, level); | |
979 | q->near_ev_cache[level] = event ? event : | |
980 | PSCHED_TICKS_PER_SEC; | |
1da177e4 | 981 | } else |
fb983d45 PM |
982 | event = q->near_ev_cache[level]; |
983 | ||
984 | if (event && next_event > event) | |
985 | next_event = event; | |
87990467 | 986 | |
1da177e4 LT |
987 | m = ~q->row_mask[level]; |
988 | while (m != (int)(-1)) { | |
87990467 | 989 | int prio = ffz(m); |
1da177e4 | 990 | m |= 1 << prio; |
87990467 | 991 | skb = htb_dequeue_tree(q, prio, level); |
1da177e4 LT |
992 | if (likely(skb != NULL)) { |
993 | sch->q.qlen--; | |
994 | sch->flags &= ~TCQ_F_THROTTLED; | |
995 | goto fin; | |
996 | } | |
997 | } | |
998 | } | |
fb983d45 PM |
999 | sch->qstats.overlimits++; |
1000 | qdisc_watchdog_schedule(&q->watchdog, next_event); | |
1da177e4 | 1001 | fin: |
1da177e4 LT |
1002 | return skb; |
1003 | } | |
1004 | ||
1005 | /* try to drop from each class (by prio) until one succeed */ | |
87990467 | 1006 | static unsigned int htb_drop(struct Qdisc *sch) |
1da177e4 LT |
1007 | { |
1008 | struct htb_sched *q = qdisc_priv(sch); | |
1009 | int prio; | |
1010 | ||
1011 | for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) { | |
1012 | struct list_head *p; | |
87990467 | 1013 | list_for_each(p, q->drops + prio) { |
1da177e4 LT |
1014 | struct htb_class *cl = list_entry(p, struct htb_class, |
1015 | un.leaf.drop_list); | |
1016 | unsigned int len; | |
87990467 SH |
1017 | if (cl->un.leaf.q->ops->drop && |
1018 | (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) { | |
1da177e4 LT |
1019 | sch->q.qlen--; |
1020 | if (!cl->un.leaf.q->q.qlen) | |
87990467 | 1021 | htb_deactivate(q, cl); |
1da177e4 LT |
1022 | return len; |
1023 | } | |
1024 | } | |
1025 | } | |
1026 | return 0; | |
1027 | } | |
1028 | ||
1029 | /* reset all classes */ | |
1030 | /* always caled under BH & queue lock */ | |
87990467 | 1031 | static void htb_reset(struct Qdisc *sch) |
1da177e4 LT |
1032 | { |
1033 | struct htb_sched *q = qdisc_priv(sch); | |
1034 | int i; | |
1da177e4 LT |
1035 | |
1036 | for (i = 0; i < HTB_HSIZE; i++) { | |
0cef296d SH |
1037 | struct hlist_node *p; |
1038 | struct htb_class *cl; | |
1039 | ||
1040 | hlist_for_each_entry(cl, p, q->hash + i, hlist) { | |
1da177e4 | 1041 | if (cl->level) |
87990467 | 1042 | memset(&cl->un.inner, 0, sizeof(cl->un.inner)); |
1da177e4 | 1043 | else { |
87990467 | 1044 | if (cl->un.leaf.q) |
1da177e4 LT |
1045 | qdisc_reset(cl->un.leaf.q); |
1046 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); | |
1047 | } | |
1048 | cl->prio_activity = 0; | |
1049 | cl->cmode = HTB_CAN_SEND; | |
1da177e4 LT |
1050 | |
1051 | } | |
1052 | } | |
fb983d45 | 1053 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
1054 | __skb_queue_purge(&q->direct_queue); |
1055 | sch->q.qlen = 0; | |
87990467 SH |
1056 | memset(q->row, 0, sizeof(q->row)); |
1057 | memset(q->row_mask, 0, sizeof(q->row_mask)); | |
1058 | memset(q->wait_pq, 0, sizeof(q->wait_pq)); | |
1059 | memset(q->ptr, 0, sizeof(q->ptr)); | |
1da177e4 | 1060 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 1061 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 LT |
1062 | } |
1063 | ||
1064 | static int htb_init(struct Qdisc *sch, struct rtattr *opt) | |
1065 | { | |
1066 | struct htb_sched *q = qdisc_priv(sch); | |
1067 | struct rtattr *tb[TCA_HTB_INIT]; | |
1068 | struct tc_htb_glob *gopt; | |
1069 | int i; | |
1da177e4 | 1070 | if (!opt || rtattr_parse_nested(tb, TCA_HTB_INIT, opt) || |
87990467 SH |
1071 | tb[TCA_HTB_INIT - 1] == NULL || |
1072 | RTA_PAYLOAD(tb[TCA_HTB_INIT - 1]) < sizeof(*gopt)) { | |
1da177e4 LT |
1073 | printk(KERN_ERR "HTB: hey probably you have bad tc tool ?\n"); |
1074 | return -EINVAL; | |
1075 | } | |
87990467 | 1076 | gopt = RTA_DATA(tb[TCA_HTB_INIT - 1]); |
1da177e4 | 1077 | if (gopt->version != HTB_VER >> 16) { |
87990467 SH |
1078 | printk(KERN_ERR |
1079 | "HTB: need tc/htb version %d (minor is %d), you have %d\n", | |
1080 | HTB_VER >> 16, HTB_VER & 0xffff, gopt->version); | |
1da177e4 LT |
1081 | return -EINVAL; |
1082 | } | |
1da177e4 LT |
1083 | |
1084 | INIT_LIST_HEAD(&q->root); | |
1085 | for (i = 0; i < HTB_HSIZE; i++) | |
0cef296d | 1086 | INIT_HLIST_HEAD(q->hash + i); |
1da177e4 | 1087 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
87990467 | 1088 | INIT_LIST_HEAD(q->drops + i); |
1da177e4 | 1089 | |
fb983d45 | 1090 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 LT |
1091 | skb_queue_head_init(&q->direct_queue); |
1092 | ||
1093 | q->direct_qlen = sch->dev->tx_queue_len; | |
87990467 | 1094 | if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */ |
1da177e4 | 1095 | q->direct_qlen = 2; |
1da177e4 LT |
1096 | |
1097 | #ifdef HTB_RATECM | |
1098 | init_timer(&q->rttim); | |
1099 | q->rttim.function = htb_rate_timer; | |
1100 | q->rttim.data = (unsigned long)sch; | |
1101 | q->rttim.expires = jiffies + HZ; | |
1102 | add_timer(&q->rttim); | |
1103 | #endif | |
1104 | if ((q->rate2quantum = gopt->rate2quantum) < 1) | |
1105 | q->rate2quantum = 1; | |
1106 | q->defcls = gopt->defcls; | |
1107 | ||
1108 | return 0; | |
1109 | } | |
1110 | ||
1111 | static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) | |
1112 | { | |
1113 | struct htb_sched *q = qdisc_priv(sch); | |
27a884dc | 1114 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
1115 | struct rtattr *rta; |
1116 | struct tc_htb_glob gopt; | |
9ac961ee | 1117 | spin_lock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
1118 | gopt.direct_pkts = q->direct_pkts; |
1119 | ||
1da177e4 LT |
1120 | gopt.version = HTB_VER; |
1121 | gopt.rate2quantum = q->rate2quantum; | |
1122 | gopt.defcls = q->defcls; | |
3bf72957 | 1123 | gopt.debug = 0; |
87990467 | 1124 | rta = (struct rtattr *)b; |
1da177e4 LT |
1125 | RTA_PUT(skb, TCA_OPTIONS, 0, NULL); |
1126 | RTA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt); | |
27a884dc | 1127 | rta->rta_len = skb_tail_pointer(skb) - b; |
9ac961ee | 1128 | spin_unlock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
1129 | return skb->len; |
1130 | rtattr_failure: | |
9ac961ee | 1131 | spin_unlock_bh(&sch->dev->queue_lock); |
dc5fc579 | 1132 | nlmsg_trim(skb, skb_tail_pointer(skb)); |
1da177e4 LT |
1133 | return -1; |
1134 | } | |
1135 | ||
1136 | static int htb_dump_class(struct Qdisc *sch, unsigned long arg, | |
87990467 | 1137 | struct sk_buff *skb, struct tcmsg *tcm) |
1da177e4 | 1138 | { |
87990467 | 1139 | struct htb_class *cl = (struct htb_class *)arg; |
27a884dc | 1140 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
1141 | struct rtattr *rta; |
1142 | struct tc_htb_opt opt; | |
1143 | ||
9ac961ee | 1144 | spin_lock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
1145 | tcm->tcm_parent = cl->parent ? cl->parent->classid : TC_H_ROOT; |
1146 | tcm->tcm_handle = cl->classid; | |
1147 | if (!cl->level && cl->un.leaf.q) | |
1148 | tcm->tcm_info = cl->un.leaf.q->handle; | |
1149 | ||
87990467 | 1150 | rta = (struct rtattr *)b; |
1da177e4 LT |
1151 | RTA_PUT(skb, TCA_OPTIONS, 0, NULL); |
1152 | ||
87990467 | 1153 | memset(&opt, 0, sizeof(opt)); |
1da177e4 | 1154 | |
87990467 SH |
1155 | opt.rate = cl->rate->rate; |
1156 | opt.buffer = cl->buffer; | |
1157 | opt.ceil = cl->ceil->rate; | |
1158 | opt.cbuffer = cl->cbuffer; | |
1159 | opt.quantum = cl->un.leaf.quantum; | |
1160 | opt.prio = cl->un.leaf.prio; | |
1161 | opt.level = cl->level; | |
1da177e4 | 1162 | RTA_PUT(skb, TCA_HTB_PARMS, sizeof(opt), &opt); |
27a884dc | 1163 | rta->rta_len = skb_tail_pointer(skb) - b; |
9ac961ee | 1164 | spin_unlock_bh(&sch->dev->queue_lock); |
1da177e4 LT |
1165 | return skb->len; |
1166 | rtattr_failure: | |
9ac961ee | 1167 | spin_unlock_bh(&sch->dev->queue_lock); |
dc5fc579 | 1168 | nlmsg_trim(skb, b); |
1da177e4 LT |
1169 | return -1; |
1170 | } | |
1171 | ||
1172 | static int | |
87990467 | 1173 | htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d) |
1da177e4 | 1174 | { |
87990467 | 1175 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1176 | |
1177 | #ifdef HTB_RATECM | |
87990467 SH |
1178 | cl->rate_est.bps = cl->rate_bytes / (HTB_EWMAC * HTB_HSIZE); |
1179 | cl->rate_est.pps = cl->rate_packets / (HTB_EWMAC * HTB_HSIZE); | |
1da177e4 LT |
1180 | #endif |
1181 | ||
1182 | if (!cl->level && cl->un.leaf.q) | |
1183 | cl->qstats.qlen = cl->un.leaf.q->q.qlen; | |
1184 | cl->xstats.tokens = cl->tokens; | |
1185 | cl->xstats.ctokens = cl->ctokens; | |
1186 | ||
1187 | if (gnet_stats_copy_basic(d, &cl->bstats) < 0 || | |
1188 | gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || | |
1189 | gnet_stats_copy_queue(d, &cl->qstats) < 0) | |
1190 | return -1; | |
1191 | ||
1192 | return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats)); | |
1193 | } | |
1194 | ||
1195 | static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
87990467 | 1196 | struct Qdisc **old) |
1da177e4 | 1197 | { |
87990467 | 1198 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1199 | |
1200 | if (cl && !cl->level) { | |
9f9afec4 PM |
1201 | if (new == NULL && |
1202 | (new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, | |
10297b99 | 1203 | cl->classid)) |
87990467 SH |
1204 | == NULL) |
1205 | return -ENOBUFS; | |
1da177e4 LT |
1206 | sch_tree_lock(sch); |
1207 | if ((*old = xchg(&cl->un.leaf.q, new)) != NULL) { | |
256d61b8 | 1208 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); |
1da177e4 LT |
1209 | qdisc_reset(*old); |
1210 | } | |
1211 | sch_tree_unlock(sch); | |
1212 | return 0; | |
1213 | } | |
1214 | return -ENOENT; | |
1215 | } | |
1216 | ||
87990467 | 1217 | static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg) |
1da177e4 | 1218 | { |
87990467 | 1219 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1220 | return (cl && !cl->level) ? cl->un.leaf.q : NULL; |
1221 | } | |
1222 | ||
256d61b8 PM |
1223 | static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg) |
1224 | { | |
1225 | struct htb_class *cl = (struct htb_class *)arg; | |
1226 | ||
1227 | if (cl->un.leaf.q->q.qlen == 0) | |
1228 | htb_deactivate(qdisc_priv(sch), cl); | |
1229 | } | |
1230 | ||
1da177e4 LT |
1231 | static unsigned long htb_get(struct Qdisc *sch, u32 classid) |
1232 | { | |
87990467 SH |
1233 | struct htb_class *cl = htb_find(classid, sch); |
1234 | if (cl) | |
1da177e4 LT |
1235 | cl->refcnt++; |
1236 | return (unsigned long)cl; | |
1237 | } | |
1238 | ||
1239 | static void htb_destroy_filters(struct tcf_proto **fl) | |
1240 | { | |
1241 | struct tcf_proto *tp; | |
1242 | ||
1243 | while ((tp = *fl) != NULL) { | |
1244 | *fl = tp->next; | |
1245 | tcf_destroy(tp); | |
1246 | } | |
1247 | } | |
1248 | ||
160d5e10 JP |
1249 | static inline int htb_parent_last_child(struct htb_class *cl) |
1250 | { | |
1251 | if (!cl->parent) | |
1252 | /* the root class */ | |
1253 | return 0; | |
1254 | ||
1255 | if (!(cl->parent->children.next == &cl->sibling && | |
1256 | cl->parent->children.prev == &cl->sibling)) | |
1257 | /* not the last child */ | |
1258 | return 0; | |
1259 | ||
1260 | return 1; | |
1261 | } | |
1262 | ||
1263 | static void htb_parent_to_leaf(struct htb_class *cl, struct Qdisc *new_q) | |
1264 | { | |
1265 | struct htb_class *parent = cl->parent; | |
1266 | ||
1267 | BUG_TRAP(!cl->level && cl->un.leaf.q && !cl->prio_activity); | |
1268 | ||
1269 | parent->level = 0; | |
1270 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1271 | INIT_LIST_HEAD(&parent->un.leaf.drop_list); | |
1272 | parent->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
1273 | parent->un.leaf.quantum = parent->quantum; | |
1274 | parent->un.leaf.prio = parent->prio; | |
1275 | parent->tokens = parent->buffer; | |
1276 | parent->ctokens = parent->cbuffer; | |
1277 | PSCHED_GET_TIME(parent->t_c); | |
1278 | parent->cmode = HTB_CAN_SEND; | |
1279 | } | |
1280 | ||
87990467 | 1281 | static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl) |
1da177e4 LT |
1282 | { |
1283 | struct htb_sched *q = qdisc_priv(sch); | |
814a175e | 1284 | |
1da177e4 LT |
1285 | if (!cl->level) { |
1286 | BUG_TRAP(cl->un.leaf.q); | |
1da177e4 LT |
1287 | qdisc_destroy(cl->un.leaf.q); |
1288 | } | |
1289 | qdisc_put_rtab(cl->rate); | |
1290 | qdisc_put_rtab(cl->ceil); | |
87990467 SH |
1291 | |
1292 | htb_destroy_filters(&cl->filter_list); | |
1293 | ||
1294 | while (!list_empty(&cl->children)) | |
1295 | htb_destroy_class(sch, list_entry(cl->children.next, | |
1296 | struct htb_class, sibling)); | |
1da177e4 LT |
1297 | |
1298 | /* note: this delete may happen twice (see htb_delete) */ | |
da33e3eb | 1299 | hlist_del_init(&cl->hlist); |
1da177e4 | 1300 | list_del(&cl->sibling); |
87990467 | 1301 | |
1da177e4 | 1302 | if (cl->prio_activity) |
87990467 SH |
1303 | htb_deactivate(q, cl); |
1304 | ||
1da177e4 | 1305 | if (cl->cmode != HTB_CAN_SEND) |
3696f625 | 1306 | htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level); |
87990467 | 1307 | |
1da177e4 LT |
1308 | kfree(cl); |
1309 | } | |
1310 | ||
1311 | /* always caled under BH & queue lock */ | |
87990467 | 1312 | static void htb_destroy(struct Qdisc *sch) |
1da177e4 LT |
1313 | { |
1314 | struct htb_sched *q = qdisc_priv(sch); | |
1da177e4 | 1315 | |
fb983d45 | 1316 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 | 1317 | #ifdef HTB_RATECM |
87990467 | 1318 | del_timer_sync(&q->rttim); |
1da177e4 LT |
1319 | #endif |
1320 | /* This line used to be after htb_destroy_class call below | |
10297b99 | 1321 | and surprisingly it worked in 2.4. But it must precede it |
1da177e4 LT |
1322 | because filter need its target class alive to be able to call |
1323 | unbind_filter on it (without Oops). */ | |
1324 | htb_destroy_filters(&q->filter_list); | |
87990467 SH |
1325 | |
1326 | while (!list_empty(&q->root)) | |
1327 | htb_destroy_class(sch, list_entry(q->root.next, | |
1328 | struct htb_class, sibling)); | |
1da177e4 LT |
1329 | |
1330 | __skb_queue_purge(&q->direct_queue); | |
1331 | } | |
1332 | ||
1333 | static int htb_delete(struct Qdisc *sch, unsigned long arg) | |
1334 | { | |
1335 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1336 | struct htb_class *cl = (struct htb_class *)arg; |
256d61b8 | 1337 | unsigned int qlen; |
160d5e10 JP |
1338 | struct Qdisc *new_q = NULL; |
1339 | int last_child = 0; | |
1da177e4 LT |
1340 | |
1341 | // TODO: why don't allow to delete subtree ? references ? does | |
1342 | // tc subsys quarantee us that in htb_destroy it holds no class | |
1343 | // refs so that we can remove children safely there ? | |
1344 | if (!list_empty(&cl->children) || cl->filter_cnt) | |
1345 | return -EBUSY; | |
87990467 | 1346 | |
160d5e10 JP |
1347 | if (!cl->level && htb_parent_last_child(cl)) { |
1348 | new_q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, | |
1349 | cl->parent->classid); | |
1350 | last_child = 1; | |
1351 | } | |
1352 | ||
1da177e4 | 1353 | sch_tree_lock(sch); |
87990467 | 1354 | |
814a175e | 1355 | if (!cl->level) { |
256d61b8 | 1356 | qlen = cl->un.leaf.q->q.qlen; |
814a175e | 1357 | qdisc_reset(cl->un.leaf.q); |
256d61b8 | 1358 | qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen); |
814a175e PM |
1359 | } |
1360 | ||
c38c83cb PM |
1361 | /* delete from hash and active; remainder in destroy_class */ |
1362 | hlist_del_init(&cl->hlist); | |
1363 | ||
1da177e4 | 1364 | if (cl->prio_activity) |
87990467 | 1365 | htb_deactivate(q, cl); |
1da177e4 | 1366 | |
160d5e10 JP |
1367 | if (last_child) |
1368 | htb_parent_to_leaf(cl, new_q); | |
1369 | ||
1da177e4 | 1370 | if (--cl->refcnt == 0) |
87990467 | 1371 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1372 | |
1373 | sch_tree_unlock(sch); | |
1374 | return 0; | |
1375 | } | |
1376 | ||
1377 | static void htb_put(struct Qdisc *sch, unsigned long arg) | |
1378 | { | |
87990467 | 1379 | struct htb_class *cl = (struct htb_class *)arg; |
1da177e4 LT |
1380 | |
1381 | if (--cl->refcnt == 0) | |
87990467 | 1382 | htb_destroy_class(sch, cl); |
1da177e4 LT |
1383 | } |
1384 | ||
87990467 SH |
1385 | static int htb_change_class(struct Qdisc *sch, u32 classid, |
1386 | u32 parentid, struct rtattr **tca, | |
1387 | unsigned long *arg) | |
1da177e4 LT |
1388 | { |
1389 | int err = -EINVAL; | |
1390 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 SH |
1391 | struct htb_class *cl = (struct htb_class *)*arg, *parent; |
1392 | struct rtattr *opt = tca[TCA_OPTIONS - 1]; | |
1da177e4 LT |
1393 | struct qdisc_rate_table *rtab = NULL, *ctab = NULL; |
1394 | struct rtattr *tb[TCA_HTB_RTAB]; | |
1395 | struct tc_htb_opt *hopt; | |
1396 | ||
1397 | /* extract all subattrs from opt attr */ | |
1398 | if (!opt || rtattr_parse_nested(tb, TCA_HTB_RTAB, opt) || | |
87990467 SH |
1399 | tb[TCA_HTB_PARMS - 1] == NULL || |
1400 | RTA_PAYLOAD(tb[TCA_HTB_PARMS - 1]) < sizeof(*hopt)) | |
1da177e4 | 1401 | goto failure; |
1da177e4 | 1402 | |
87990467 SH |
1403 | parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch); |
1404 | ||
1405 | hopt = RTA_DATA(tb[TCA_HTB_PARMS - 1]); | |
3bf72957 | 1406 | |
87990467 SH |
1407 | rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB - 1]); |
1408 | ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB - 1]); | |
1409 | if (!rtab || !ctab) | |
1410 | goto failure; | |
1da177e4 | 1411 | |
87990467 | 1412 | if (!cl) { /* new class */ |
1da177e4 | 1413 | struct Qdisc *new_q; |
3696f625 SH |
1414 | int prio; |
1415 | ||
1da177e4 | 1416 | /* check for valid classid */ |
87990467 SH |
1417 | if (!classid || TC_H_MAJ(classid ^ sch->handle) |
1418 | || htb_find(classid, sch)) | |
1da177e4 LT |
1419 | goto failure; |
1420 | ||
1421 | /* check maximal depth */ | |
1422 | if (parent && parent->parent && parent->parent->level < 2) { | |
1423 | printk(KERN_ERR "htb: tree is too deep\n"); | |
1424 | goto failure; | |
1425 | } | |
1426 | err = -ENOBUFS; | |
0da974f4 | 1427 | if ((cl = kzalloc(sizeof(*cl), GFP_KERNEL)) == NULL) |
1da177e4 | 1428 | goto failure; |
87990467 | 1429 | |
1da177e4 LT |
1430 | cl->refcnt = 1; |
1431 | INIT_LIST_HEAD(&cl->sibling); | |
0cef296d | 1432 | INIT_HLIST_NODE(&cl->hlist); |
1da177e4 LT |
1433 | INIT_LIST_HEAD(&cl->children); |
1434 | INIT_LIST_HEAD(&cl->un.leaf.drop_list); | |
3696f625 SH |
1435 | RB_CLEAR_NODE(&cl->pq_node); |
1436 | ||
1437 | for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) | |
1438 | RB_CLEAR_NODE(&cl->node[prio]); | |
1da177e4 LT |
1439 | |
1440 | /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL) | |
1441 | so that can't be used inside of sch_tree_lock | |
1442 | -- thanks to Karlis Peisenieks */ | |
9f9afec4 | 1443 | new_q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, classid); |
1da177e4 LT |
1444 | sch_tree_lock(sch); |
1445 | if (parent && !parent->level) { | |
256d61b8 PM |
1446 | unsigned int qlen = parent->un.leaf.q->q.qlen; |
1447 | ||
1da177e4 | 1448 | /* turn parent into inner node */ |
256d61b8 PM |
1449 | qdisc_reset(parent->un.leaf.q); |
1450 | qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen); | |
87990467 SH |
1451 | qdisc_destroy(parent->un.leaf.q); |
1452 | if (parent->prio_activity) | |
1453 | htb_deactivate(q, parent); | |
1da177e4 LT |
1454 | |
1455 | /* remove from evt list because of level change */ | |
1456 | if (parent->cmode != HTB_CAN_SEND) { | |
3696f625 | 1457 | htb_safe_rb_erase(&parent->pq_node, q->wait_pq); |
1da177e4 LT |
1458 | parent->cmode = HTB_CAN_SEND; |
1459 | } | |
1460 | parent->level = (parent->parent ? parent->parent->level | |
87990467 SH |
1461 | : TC_HTB_MAXDEPTH) - 1; |
1462 | memset(&parent->un.inner, 0, sizeof(parent->un.inner)); | |
1da177e4 LT |
1463 | } |
1464 | /* leaf (we) needs elementary qdisc */ | |
1465 | cl->un.leaf.q = new_q ? new_q : &noop_qdisc; | |
1466 | ||
87990467 SH |
1467 | cl->classid = classid; |
1468 | cl->parent = parent; | |
1da177e4 LT |
1469 | |
1470 | /* set class to be in HTB_CAN_SEND state */ | |
1471 | cl->tokens = hopt->buffer; | |
1472 | cl->ctokens = hopt->cbuffer; | |
00c04af9 | 1473 | cl->mbuffer = 60 * PSCHED_TICKS_PER_SEC; /* 1min */ |
1da177e4 LT |
1474 | PSCHED_GET_TIME(cl->t_c); |
1475 | cl->cmode = HTB_CAN_SEND; | |
1476 | ||
1477 | /* attach to the hash list and parent's family */ | |
0cef296d | 1478 | hlist_add_head(&cl->hlist, q->hash + htb_hash(classid)); |
87990467 SH |
1479 | list_add_tail(&cl->sibling, |
1480 | parent ? &parent->children : &q->root); | |
1481 | } else | |
1482 | sch_tree_lock(sch); | |
1da177e4 LT |
1483 | |
1484 | /* it used to be a nasty bug here, we have to check that node | |
87990467 | 1485 | is really leaf before changing cl->un.leaf ! */ |
1da177e4 LT |
1486 | if (!cl->level) { |
1487 | cl->un.leaf.quantum = rtab->rate.rate / q->rate2quantum; | |
1488 | if (!hopt->quantum && cl->un.leaf.quantum < 1000) { | |
87990467 SH |
1489 | printk(KERN_WARNING |
1490 | "HTB: quantum of class %X is small. Consider r2q change.\n", | |
1491 | cl->classid); | |
1da177e4 LT |
1492 | cl->un.leaf.quantum = 1000; |
1493 | } | |
1494 | if (!hopt->quantum && cl->un.leaf.quantum > 200000) { | |
87990467 SH |
1495 | printk(KERN_WARNING |
1496 | "HTB: quantum of class %X is big. Consider r2q change.\n", | |
1497 | cl->classid); | |
1da177e4 LT |
1498 | cl->un.leaf.quantum = 200000; |
1499 | } | |
1500 | if (hopt->quantum) | |
1501 | cl->un.leaf.quantum = hopt->quantum; | |
1502 | if ((cl->un.leaf.prio = hopt->prio) >= TC_HTB_NUMPRIO) | |
1503 | cl->un.leaf.prio = TC_HTB_NUMPRIO - 1; | |
160d5e10 JP |
1504 | |
1505 | /* backup for htb_parent_to_leaf */ | |
1506 | cl->quantum = cl->un.leaf.quantum; | |
1507 | cl->prio = cl->un.leaf.prio; | |
1da177e4 LT |
1508 | } |
1509 | ||
1510 | cl->buffer = hopt->buffer; | |
1511 | cl->cbuffer = hopt->cbuffer; | |
87990467 SH |
1512 | if (cl->rate) |
1513 | qdisc_put_rtab(cl->rate); | |
1514 | cl->rate = rtab; | |
1515 | if (cl->ceil) | |
1516 | qdisc_put_rtab(cl->ceil); | |
1517 | cl->ceil = ctab; | |
1da177e4 LT |
1518 | sch_tree_unlock(sch); |
1519 | ||
1520 | *arg = (unsigned long)cl; | |
1521 | return 0; | |
1522 | ||
1523 | failure: | |
87990467 SH |
1524 | if (rtab) |
1525 | qdisc_put_rtab(rtab); | |
1526 | if (ctab) | |
1527 | qdisc_put_rtab(ctab); | |
1da177e4 LT |
1528 | return err; |
1529 | } | |
1530 | ||
1531 | static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg) | |
1532 | { | |
1533 | struct htb_sched *q = qdisc_priv(sch); | |
1534 | struct htb_class *cl = (struct htb_class *)arg; | |
1535 | struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list; | |
3bf72957 | 1536 | |
1da177e4 LT |
1537 | return fl; |
1538 | } | |
1539 | ||
1540 | static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, | |
87990467 | 1541 | u32 classid) |
1da177e4 LT |
1542 | { |
1543 | struct htb_sched *q = qdisc_priv(sch); | |
87990467 | 1544 | struct htb_class *cl = htb_find(classid, sch); |
3bf72957 | 1545 | |
1da177e4 | 1546 | /*if (cl && !cl->level) return 0; |
87990467 SH |
1547 | The line above used to be there to prevent attaching filters to |
1548 | leaves. But at least tc_index filter uses this just to get class | |
1549 | for other reasons so that we have to allow for it. | |
1550 | ---- | |
1551 | 19.6.2002 As Werner explained it is ok - bind filter is just | |
1552 | another way to "lock" the class - unlike "get" this lock can | |
1553 | be broken by class during destroy IIUC. | |
1da177e4 | 1554 | */ |
87990467 SH |
1555 | if (cl) |
1556 | cl->filter_cnt++; | |
1557 | else | |
1da177e4 LT |
1558 | q->filter_cnt++; |
1559 | return (unsigned long)cl; | |
1560 | } | |
1561 | ||
1562 | static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) | |
1563 | { | |
1564 | struct htb_sched *q = qdisc_priv(sch); | |
1565 | struct htb_class *cl = (struct htb_class *)arg; | |
3bf72957 | 1566 | |
87990467 SH |
1567 | if (cl) |
1568 | cl->filter_cnt--; | |
1569 | else | |
1da177e4 LT |
1570 | q->filter_cnt--; |
1571 | } | |
1572 | ||
1573 | static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
1574 | { | |
1575 | struct htb_sched *q = qdisc_priv(sch); | |
1576 | int i; | |
1577 | ||
1578 | if (arg->stop) | |
1579 | return; | |
1580 | ||
1581 | for (i = 0; i < HTB_HSIZE; i++) { | |
0cef296d SH |
1582 | struct hlist_node *p; |
1583 | struct htb_class *cl; | |
1584 | ||
1585 | hlist_for_each_entry(cl, p, q->hash + i, hlist) { | |
1da177e4 LT |
1586 | if (arg->count < arg->skip) { |
1587 | arg->count++; | |
1588 | continue; | |
1589 | } | |
1590 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
1591 | arg->stop = 1; | |
1592 | return; | |
1593 | } | |
1594 | arg->count++; | |
1595 | } | |
1596 | } | |
1597 | } | |
1598 | ||
1599 | static struct Qdisc_class_ops htb_class_ops = { | |
1600 | .graft = htb_graft, | |
1601 | .leaf = htb_leaf, | |
256d61b8 | 1602 | .qlen_notify = htb_qlen_notify, |
1da177e4 LT |
1603 | .get = htb_get, |
1604 | .put = htb_put, | |
1605 | .change = htb_change_class, | |
1606 | .delete = htb_delete, | |
1607 | .walk = htb_walk, | |
1608 | .tcf_chain = htb_find_tcf, | |
1609 | .bind_tcf = htb_bind_filter, | |
1610 | .unbind_tcf = htb_unbind_filter, | |
1611 | .dump = htb_dump_class, | |
1612 | .dump_stats = htb_dump_class_stats, | |
1613 | }; | |
1614 | ||
1615 | static struct Qdisc_ops htb_qdisc_ops = { | |
1616 | .next = NULL, | |
1617 | .cl_ops = &htb_class_ops, | |
1618 | .id = "htb", | |
1619 | .priv_size = sizeof(struct htb_sched), | |
1620 | .enqueue = htb_enqueue, | |
1621 | .dequeue = htb_dequeue, | |
1622 | .requeue = htb_requeue, | |
1623 | .drop = htb_drop, | |
1624 | .init = htb_init, | |
1625 | .reset = htb_reset, | |
1626 | .destroy = htb_destroy, | |
1627 | .change = NULL /* htb_change */, | |
1628 | .dump = htb_dump, | |
1629 | .owner = THIS_MODULE, | |
1630 | }; | |
1631 | ||
1632 | static int __init htb_module_init(void) | |
1633 | { | |
87990467 | 1634 | return register_qdisc(&htb_qdisc_ops); |
1da177e4 | 1635 | } |
87990467 | 1636 | static void __exit htb_module_exit(void) |
1da177e4 | 1637 | { |
87990467 | 1638 | unregister_qdisc(&htb_qdisc_ops); |
1da177e4 | 1639 | } |
87990467 | 1640 | |
1da177e4 LT |
1641 | module_init(htb_module_init) |
1642 | module_exit(htb_module_exit) | |
1643 | MODULE_LICENSE("GPL"); |