Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * net/sched/sch_api.c Packet scheduler API. | |
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: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> | |
10 | * | |
11 | * Fixes: | |
12 | * | |
13 | * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired. | |
14 | * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support | |
15 | * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support | |
16 | */ | |
17 | ||
1da177e4 LT |
18 | #include <linux/module.h> |
19 | #include <linux/types.h> | |
20 | #include <linux/kernel.h> | |
1da177e4 LT |
21 | #include <linux/string.h> |
22 | #include <linux/mm.h> | |
23 | #include <linux/socket.h> | |
24 | #include <linux/sockios.h> | |
25 | #include <linux/in.h> | |
26 | #include <linux/errno.h> | |
27 | #include <linux/interrupt.h> | |
28 | #include <linux/netdevice.h> | |
29 | #include <linux/skbuff.h> | |
30 | #include <linux/rtnetlink.h> | |
31 | #include <linux/init.h> | |
32 | #include <linux/proc_fs.h> | |
33 | #include <linux/seq_file.h> | |
34 | #include <linux/kmod.h> | |
35 | #include <linux/list.h> | |
36 | #include <linux/bitops.h> | |
4179477f | 37 | #include <linux/hrtimer.h> |
1da177e4 | 38 | |
dc5fc579 | 39 | #include <net/netlink.h> |
1da177e4 LT |
40 | #include <net/sock.h> |
41 | #include <net/pkt_sched.h> | |
42 | ||
43 | #include <asm/processor.h> | |
44 | #include <asm/uaccess.h> | |
45 | #include <asm/system.h> | |
46 | ||
47 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid, | |
48 | struct Qdisc *old, struct Qdisc *new); | |
49 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
50 | struct Qdisc *q, unsigned long cl, int event); | |
51 | ||
52 | /* | |
53 | ||
54 | Short review. | |
55 | ------------- | |
56 | ||
57 | This file consists of two interrelated parts: | |
58 | ||
59 | 1. queueing disciplines manager frontend. | |
60 | 2. traffic classes manager frontend. | |
61 | ||
62 | Generally, queueing discipline ("qdisc") is a black box, | |
63 | which is able to enqueue packets and to dequeue them (when | |
64 | device is ready to send something) in order and at times | |
65 | determined by algorithm hidden in it. | |
66 | ||
67 | qdisc's are divided to two categories: | |
68 | - "queues", which have no internal structure visible from outside. | |
69 | - "schedulers", which split all the packets to "traffic classes", | |
70 | using "packet classifiers" (look at cls_api.c) | |
71 | ||
72 | In turn, classes may have child qdiscs (as rule, queues) | |
73 | attached to them etc. etc. etc. | |
74 | ||
75 | The goal of the routines in this file is to translate | |
76 | information supplied by user in the form of handles | |
77 | to more intelligible for kernel form, to make some sanity | |
78 | checks and part of work, which is common to all qdiscs | |
79 | and to provide rtnetlink notifications. | |
80 | ||
81 | All real intelligent work is done inside qdisc modules. | |
82 | ||
83 | ||
84 | ||
85 | Every discipline has two major routines: enqueue and dequeue. | |
86 | ||
87 | ---dequeue | |
88 | ||
89 | dequeue usually returns a skb to send. It is allowed to return NULL, | |
90 | but it does not mean that queue is empty, it just means that | |
91 | discipline does not want to send anything this time. | |
92 | Queue is really empty if q->q.qlen == 0. | |
93 | For complicated disciplines with multiple queues q->q is not | |
94 | real packet queue, but however q->q.qlen must be valid. | |
95 | ||
96 | ---enqueue | |
97 | ||
98 | enqueue returns 0, if packet was enqueued successfully. | |
99 | If packet (this one or another one) was dropped, it returns | |
100 | not zero error code. | |
101 | NET_XMIT_DROP - this packet dropped | |
102 | Expected action: do not backoff, but wait until queue will clear. | |
103 | NET_XMIT_CN - probably this packet enqueued, but another one dropped. | |
104 | Expected action: backoff or ignore | |
105 | NET_XMIT_POLICED - dropped by police. | |
106 | Expected action: backoff or error to real-time apps. | |
107 | ||
108 | Auxiliary routines: | |
109 | ||
110 | ---requeue | |
111 | ||
112 | requeues once dequeued packet. It is used for non-standard or | |
113 | just buggy devices, which can defer output even if dev->tbusy=0. | |
114 | ||
115 | ---reset | |
116 | ||
117 | returns qdisc to initial state: purge all buffers, clear all | |
118 | timers, counters (except for statistics) etc. | |
119 | ||
120 | ---init | |
121 | ||
122 | initializes newly created qdisc. | |
123 | ||
124 | ---destroy | |
125 | ||
126 | destroys resources allocated by init and during lifetime of qdisc. | |
127 | ||
128 | ---change | |
129 | ||
130 | changes qdisc parameters. | |
131 | */ | |
132 | ||
133 | /* Protects list of registered TC modules. It is pure SMP lock. */ | |
134 | static DEFINE_RWLOCK(qdisc_mod_lock); | |
135 | ||
136 | ||
137 | /************************************************ | |
138 | * Queueing disciplines manipulation. * | |
139 | ************************************************/ | |
140 | ||
141 | ||
142 | /* The list of all installed queueing disciplines. */ | |
143 | ||
144 | static struct Qdisc_ops *qdisc_base; | |
145 | ||
146 | /* Register/uregister queueing discipline */ | |
147 | ||
148 | int register_qdisc(struct Qdisc_ops *qops) | |
149 | { | |
150 | struct Qdisc_ops *q, **qp; | |
151 | int rc = -EEXIST; | |
152 | ||
153 | write_lock(&qdisc_mod_lock); | |
154 | for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) | |
155 | if (!strcmp(qops->id, q->id)) | |
156 | goto out; | |
157 | ||
158 | if (qops->enqueue == NULL) | |
159 | qops->enqueue = noop_qdisc_ops.enqueue; | |
160 | if (qops->requeue == NULL) | |
161 | qops->requeue = noop_qdisc_ops.requeue; | |
162 | if (qops->dequeue == NULL) | |
163 | qops->dequeue = noop_qdisc_ops.dequeue; | |
164 | ||
165 | qops->next = NULL; | |
166 | *qp = qops; | |
167 | rc = 0; | |
168 | out: | |
169 | write_unlock(&qdisc_mod_lock); | |
170 | return rc; | |
171 | } | |
172 | ||
173 | int unregister_qdisc(struct Qdisc_ops *qops) | |
174 | { | |
175 | struct Qdisc_ops *q, **qp; | |
176 | int err = -ENOENT; | |
177 | ||
178 | write_lock(&qdisc_mod_lock); | |
179 | for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next) | |
180 | if (q == qops) | |
181 | break; | |
182 | if (q) { | |
183 | *qp = q->next; | |
184 | q->next = NULL; | |
185 | err = 0; | |
186 | } | |
187 | write_unlock(&qdisc_mod_lock); | |
188 | return err; | |
189 | } | |
190 | ||
191 | /* We know handle. Find qdisc among all qdisc's attached to device | |
192 | (root qdisc, all its children, children of children etc.) | |
193 | */ | |
194 | ||
43effa1e | 195 | static struct Qdisc *__qdisc_lookup(struct net_device *dev, u32 handle) |
1da177e4 LT |
196 | { |
197 | struct Qdisc *q; | |
198 | ||
1da177e4 | 199 | list_for_each_entry(q, &dev->qdisc_list, list) { |
43effa1e | 200 | if (q->handle == handle) |
1da177e4 | 201 | return q; |
1da177e4 | 202 | } |
1da177e4 LT |
203 | return NULL; |
204 | } | |
205 | ||
43effa1e PM |
206 | struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) |
207 | { | |
208 | struct Qdisc *q; | |
209 | ||
210 | read_lock(&qdisc_tree_lock); | |
211 | q = __qdisc_lookup(dev, handle); | |
212 | read_unlock(&qdisc_tree_lock); | |
213 | return q; | |
214 | } | |
215 | ||
1da177e4 LT |
216 | static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) |
217 | { | |
218 | unsigned long cl; | |
219 | struct Qdisc *leaf; | |
220 | struct Qdisc_class_ops *cops = p->ops->cl_ops; | |
221 | ||
222 | if (cops == NULL) | |
223 | return NULL; | |
224 | cl = cops->get(p, classid); | |
225 | ||
226 | if (cl == 0) | |
227 | return NULL; | |
228 | leaf = cops->leaf(p, cl); | |
229 | cops->put(p, cl); | |
230 | return leaf; | |
231 | } | |
232 | ||
233 | /* Find queueing discipline by name */ | |
234 | ||
235 | static struct Qdisc_ops *qdisc_lookup_ops(struct rtattr *kind) | |
236 | { | |
237 | struct Qdisc_ops *q = NULL; | |
238 | ||
239 | if (kind) { | |
240 | read_lock(&qdisc_mod_lock); | |
241 | for (q = qdisc_base; q; q = q->next) { | |
242 | if (rtattr_strcmp(kind, q->id) == 0) { | |
243 | if (!try_module_get(q->owner)) | |
244 | q = NULL; | |
245 | break; | |
246 | } | |
247 | } | |
248 | read_unlock(&qdisc_mod_lock); | |
249 | } | |
250 | return q; | |
251 | } | |
252 | ||
253 | static struct qdisc_rate_table *qdisc_rtab_list; | |
254 | ||
255 | struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct rtattr *tab) | |
256 | { | |
257 | struct qdisc_rate_table *rtab; | |
258 | ||
259 | for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { | |
260 | if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) { | |
261 | rtab->refcnt++; | |
262 | return rtab; | |
263 | } | |
264 | } | |
265 | ||
266 | if (tab == NULL || r->rate == 0 || r->cell_log == 0 || RTA_PAYLOAD(tab) != 1024) | |
267 | return NULL; | |
268 | ||
269 | rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); | |
270 | if (rtab) { | |
271 | rtab->rate = *r; | |
272 | rtab->refcnt = 1; | |
273 | memcpy(rtab->data, RTA_DATA(tab), 1024); | |
274 | rtab->next = qdisc_rtab_list; | |
275 | qdisc_rtab_list = rtab; | |
276 | } | |
277 | return rtab; | |
278 | } | |
279 | ||
280 | void qdisc_put_rtab(struct qdisc_rate_table *tab) | |
281 | { | |
282 | struct qdisc_rate_table *rtab, **rtabp; | |
283 | ||
284 | if (!tab || --tab->refcnt) | |
285 | return; | |
286 | ||
287 | for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) { | |
288 | if (rtab == tab) { | |
289 | *rtabp = rtab->next; | |
290 | kfree(rtab); | |
291 | return; | |
292 | } | |
293 | } | |
294 | } | |
295 | ||
4179477f PM |
296 | static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) |
297 | { | |
298 | struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, | |
299 | timer); | |
300 | ||
301 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
302 | netif_schedule(wd->qdisc->dev); | |
303 | return HRTIMER_NORESTART; | |
304 | } | |
305 | ||
306 | void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) | |
307 | { | |
308 | hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
309 | wd->timer.function = qdisc_watchdog; | |
310 | wd->qdisc = qdisc; | |
311 | } | |
312 | EXPORT_SYMBOL(qdisc_watchdog_init); | |
313 | ||
314 | void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires) | |
315 | { | |
316 | ktime_t time; | |
317 | ||
318 | wd->qdisc->flags |= TCQ_F_THROTTLED; | |
319 | time = ktime_set(0, 0); | |
320 | time = ktime_add_ns(time, PSCHED_US2NS(expires)); | |
321 | hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS); | |
322 | } | |
323 | EXPORT_SYMBOL(qdisc_watchdog_schedule); | |
324 | ||
325 | void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) | |
326 | { | |
327 | hrtimer_cancel(&wd->timer); | |
328 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
329 | } | |
330 | EXPORT_SYMBOL(qdisc_watchdog_cancel); | |
1da177e4 LT |
331 | |
332 | /* Allocate an unique handle from space managed by kernel */ | |
333 | ||
334 | static u32 qdisc_alloc_handle(struct net_device *dev) | |
335 | { | |
336 | int i = 0x10000; | |
337 | static u32 autohandle = TC_H_MAKE(0x80000000U, 0); | |
338 | ||
339 | do { | |
340 | autohandle += TC_H_MAKE(0x10000U, 0); | |
341 | if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) | |
342 | autohandle = TC_H_MAKE(0x80000000U, 0); | |
343 | } while (qdisc_lookup(dev, autohandle) && --i > 0); | |
344 | ||
345 | return i>0 ? autohandle : 0; | |
346 | } | |
347 | ||
348 | /* Attach toplevel qdisc to device dev */ | |
349 | ||
350 | static struct Qdisc * | |
351 | dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc) | |
352 | { | |
353 | struct Qdisc *oqdisc; | |
354 | ||
355 | if (dev->flags & IFF_UP) | |
356 | dev_deactivate(dev); | |
357 | ||
358 | qdisc_lock_tree(dev); | |
359 | if (qdisc && qdisc->flags&TCQ_F_INGRESS) { | |
360 | oqdisc = dev->qdisc_ingress; | |
361 | /* Prune old scheduler */ | |
362 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) { | |
363 | /* delete */ | |
364 | qdisc_reset(oqdisc); | |
365 | dev->qdisc_ingress = NULL; | |
366 | } else { /* new */ | |
367 | dev->qdisc_ingress = qdisc; | |
368 | } | |
369 | ||
370 | } else { | |
371 | ||
372 | oqdisc = dev->qdisc_sleeping; | |
373 | ||
374 | /* Prune old scheduler */ | |
375 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
376 | qdisc_reset(oqdisc); | |
377 | ||
378 | /* ... and graft new one */ | |
379 | if (qdisc == NULL) | |
380 | qdisc = &noop_qdisc; | |
381 | dev->qdisc_sleeping = qdisc; | |
382 | dev->qdisc = &noop_qdisc; | |
383 | } | |
384 | ||
385 | qdisc_unlock_tree(dev); | |
386 | ||
387 | if (dev->flags & IFF_UP) | |
388 | dev_activate(dev); | |
389 | ||
390 | return oqdisc; | |
391 | } | |
392 | ||
43effa1e PM |
393 | void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n) |
394 | { | |
395 | struct Qdisc_class_ops *cops; | |
396 | unsigned long cl; | |
397 | u32 parentid; | |
398 | ||
399 | if (n == 0) | |
400 | return; | |
401 | while ((parentid = sch->parent)) { | |
402 | sch = __qdisc_lookup(sch->dev, TC_H_MAJ(parentid)); | |
403 | cops = sch->ops->cl_ops; | |
404 | if (cops->qlen_notify) { | |
405 | cl = cops->get(sch, parentid); | |
406 | cops->qlen_notify(sch, cl); | |
407 | cops->put(sch, cl); | |
408 | } | |
409 | sch->q.qlen -= n; | |
410 | } | |
411 | } | |
412 | EXPORT_SYMBOL(qdisc_tree_decrease_qlen); | |
1da177e4 LT |
413 | |
414 | /* Graft qdisc "new" to class "classid" of qdisc "parent" or | |
415 | to device "dev". | |
416 | ||
417 | Old qdisc is not destroyed but returned in *old. | |
418 | */ | |
419 | ||
420 | static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, | |
421 | u32 classid, | |
422 | struct Qdisc *new, struct Qdisc **old) | |
423 | { | |
424 | int err = 0; | |
425 | struct Qdisc *q = *old; | |
426 | ||
427 | ||
10297b99 | 428 | if (parent == NULL) { |
1da177e4 LT |
429 | if (q && q->flags&TCQ_F_INGRESS) { |
430 | *old = dev_graft_qdisc(dev, q); | |
431 | } else { | |
432 | *old = dev_graft_qdisc(dev, new); | |
433 | } | |
434 | } else { | |
435 | struct Qdisc_class_ops *cops = parent->ops->cl_ops; | |
436 | ||
437 | err = -EINVAL; | |
438 | ||
439 | if (cops) { | |
440 | unsigned long cl = cops->get(parent, classid); | |
441 | if (cl) { | |
442 | err = cops->graft(parent, cl, new, old); | |
443 | if (new) | |
444 | new->parent = classid; | |
445 | cops->put(parent, cl); | |
446 | } | |
447 | } | |
448 | } | |
449 | return err; | |
450 | } | |
451 | ||
452 | /* | |
453 | Allocate and initialize new qdisc. | |
454 | ||
455 | Parameters are passed via opt. | |
456 | */ | |
457 | ||
458 | static struct Qdisc * | |
459 | qdisc_create(struct net_device *dev, u32 handle, struct rtattr **tca, int *errp) | |
460 | { | |
461 | int err; | |
462 | struct rtattr *kind = tca[TCA_KIND-1]; | |
1da177e4 LT |
463 | struct Qdisc *sch; |
464 | struct Qdisc_ops *ops; | |
1da177e4 LT |
465 | |
466 | ops = qdisc_lookup_ops(kind); | |
467 | #ifdef CONFIG_KMOD | |
468 | if (ops == NULL && kind != NULL) { | |
469 | char name[IFNAMSIZ]; | |
470 | if (rtattr_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { | |
471 | /* We dropped the RTNL semaphore in order to | |
472 | * perform the module load. So, even if we | |
473 | * succeeded in loading the module we have to | |
474 | * tell the caller to replay the request. We | |
475 | * indicate this using -EAGAIN. | |
476 | * We replay the request because the device may | |
477 | * go away in the mean time. | |
478 | */ | |
479 | rtnl_unlock(); | |
480 | request_module("sch_%s", name); | |
481 | rtnl_lock(); | |
482 | ops = qdisc_lookup_ops(kind); | |
483 | if (ops != NULL) { | |
484 | /* We will try again qdisc_lookup_ops, | |
485 | * so don't keep a reference. | |
486 | */ | |
487 | module_put(ops->owner); | |
488 | err = -EAGAIN; | |
489 | goto err_out; | |
490 | } | |
491 | } | |
492 | } | |
493 | #endif | |
494 | ||
b9e2cc0f | 495 | err = -ENOENT; |
1da177e4 LT |
496 | if (ops == NULL) |
497 | goto err_out; | |
498 | ||
3d54b82f TG |
499 | sch = qdisc_alloc(dev, ops); |
500 | if (IS_ERR(sch)) { | |
501 | err = PTR_ERR(sch); | |
1da177e4 | 502 | goto err_out2; |
3d54b82f | 503 | } |
1da177e4 | 504 | |
3d54b82f | 505 | if (handle == TC_H_INGRESS) { |
1da177e4 | 506 | sch->flags |= TCQ_F_INGRESS; |
3d54b82f TG |
507 | handle = TC_H_MAKE(TC_H_INGRESS, 0); |
508 | } else if (handle == 0) { | |
1da177e4 LT |
509 | handle = qdisc_alloc_handle(dev); |
510 | err = -ENOMEM; | |
511 | if (handle == 0) | |
512 | goto err_out3; | |
513 | } | |
514 | ||
3d54b82f | 515 | sch->handle = handle; |
1da177e4 LT |
516 | |
517 | if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS-1])) == 0) { | |
023e09a7 TG |
518 | #ifdef CONFIG_NET_ESTIMATOR |
519 | if (tca[TCA_RATE-1]) { | |
520 | err = gen_new_estimator(&sch->bstats, &sch->rate_est, | |
521 | sch->stats_lock, | |
522 | tca[TCA_RATE-1]); | |
523 | if (err) { | |
524 | /* | |
525 | * Any broken qdiscs that would require | |
526 | * a ops->reset() here? The qdisc was never | |
527 | * in action so it shouldn't be necessary. | |
528 | */ | |
529 | if (ops->destroy) | |
530 | ops->destroy(sch); | |
531 | goto err_out3; | |
532 | } | |
533 | } | |
534 | #endif | |
1da177e4 LT |
535 | qdisc_lock_tree(dev); |
536 | list_add_tail(&sch->list, &dev->qdisc_list); | |
537 | qdisc_unlock_tree(dev); | |
538 | ||
1da177e4 LT |
539 | return sch; |
540 | } | |
541 | err_out3: | |
542 | dev_put(dev); | |
3d54b82f | 543 | kfree((char *) sch - sch->padded); |
1da177e4 LT |
544 | err_out2: |
545 | module_put(ops->owner); | |
546 | err_out: | |
547 | *errp = err; | |
1da177e4 LT |
548 | return NULL; |
549 | } | |
550 | ||
551 | static int qdisc_change(struct Qdisc *sch, struct rtattr **tca) | |
552 | { | |
553 | if (tca[TCA_OPTIONS-1]) { | |
554 | int err; | |
555 | ||
556 | if (sch->ops->change == NULL) | |
557 | return -EINVAL; | |
558 | err = sch->ops->change(sch, tca[TCA_OPTIONS-1]); | |
559 | if (err) | |
560 | return err; | |
561 | } | |
562 | #ifdef CONFIG_NET_ESTIMATOR | |
563 | if (tca[TCA_RATE-1]) | |
564 | gen_replace_estimator(&sch->bstats, &sch->rate_est, | |
565 | sch->stats_lock, tca[TCA_RATE-1]); | |
566 | #endif | |
567 | return 0; | |
568 | } | |
569 | ||
570 | struct check_loop_arg | |
571 | { | |
572 | struct qdisc_walker w; | |
573 | struct Qdisc *p; | |
574 | int depth; | |
575 | }; | |
576 | ||
577 | static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); | |
578 | ||
579 | static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) | |
580 | { | |
581 | struct check_loop_arg arg; | |
582 | ||
583 | if (q->ops->cl_ops == NULL) | |
584 | return 0; | |
585 | ||
586 | arg.w.stop = arg.w.skip = arg.w.count = 0; | |
587 | arg.w.fn = check_loop_fn; | |
588 | arg.depth = depth; | |
589 | arg.p = p; | |
590 | q->ops->cl_ops->walk(q, &arg.w); | |
591 | return arg.w.stop ? -ELOOP : 0; | |
592 | } | |
593 | ||
594 | static int | |
595 | check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) | |
596 | { | |
597 | struct Qdisc *leaf; | |
598 | struct Qdisc_class_ops *cops = q->ops->cl_ops; | |
599 | struct check_loop_arg *arg = (struct check_loop_arg *)w; | |
600 | ||
601 | leaf = cops->leaf(q, cl); | |
602 | if (leaf) { | |
603 | if (leaf == arg->p || arg->depth > 7) | |
604 | return -ELOOP; | |
605 | return check_loop(leaf, arg->p, arg->depth + 1); | |
606 | } | |
607 | return 0; | |
608 | } | |
609 | ||
610 | /* | |
611 | * Delete/get qdisc. | |
612 | */ | |
613 | ||
614 | static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
615 | { | |
616 | struct tcmsg *tcm = NLMSG_DATA(n); | |
617 | struct rtattr **tca = arg; | |
618 | struct net_device *dev; | |
619 | u32 clid = tcm->tcm_parent; | |
620 | struct Qdisc *q = NULL; | |
621 | struct Qdisc *p = NULL; | |
622 | int err; | |
623 | ||
624 | if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
625 | return -ENODEV; | |
626 | ||
627 | if (clid) { | |
628 | if (clid != TC_H_ROOT) { | |
629 | if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { | |
630 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
631 | return -ENOENT; | |
632 | q = qdisc_leaf(p, clid); | |
633 | } else { /* ingress */ | |
634 | q = dev->qdisc_ingress; | |
10297b99 | 635 | } |
1da177e4 LT |
636 | } else { |
637 | q = dev->qdisc_sleeping; | |
638 | } | |
639 | if (!q) | |
640 | return -ENOENT; | |
641 | ||
642 | if (tcm->tcm_handle && q->handle != tcm->tcm_handle) | |
643 | return -EINVAL; | |
644 | } else { | |
645 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
646 | return -ENOENT; | |
647 | } | |
648 | ||
649 | if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)) | |
650 | return -EINVAL; | |
651 | ||
652 | if (n->nlmsg_type == RTM_DELQDISC) { | |
653 | if (!clid) | |
654 | return -EINVAL; | |
655 | if (q->handle == 0) | |
656 | return -ENOENT; | |
657 | if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0) | |
658 | return err; | |
659 | if (q) { | |
660 | qdisc_notify(skb, n, clid, q, NULL); | |
661 | spin_lock_bh(&dev->queue_lock); | |
662 | qdisc_destroy(q); | |
663 | spin_unlock_bh(&dev->queue_lock); | |
664 | } | |
665 | } else { | |
666 | qdisc_notify(skb, n, clid, NULL, q); | |
667 | } | |
668 | return 0; | |
669 | } | |
670 | ||
671 | /* | |
672 | Create/change qdisc. | |
673 | */ | |
674 | ||
675 | static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
676 | { | |
677 | struct tcmsg *tcm; | |
678 | struct rtattr **tca; | |
679 | struct net_device *dev; | |
680 | u32 clid; | |
681 | struct Qdisc *q, *p; | |
682 | int err; | |
683 | ||
684 | replay: | |
685 | /* Reinit, just in case something touches this. */ | |
686 | tcm = NLMSG_DATA(n); | |
687 | tca = arg; | |
688 | clid = tcm->tcm_parent; | |
689 | q = p = NULL; | |
690 | ||
691 | if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
692 | return -ENODEV; | |
693 | ||
694 | if (clid) { | |
695 | if (clid != TC_H_ROOT) { | |
696 | if (clid != TC_H_INGRESS) { | |
697 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
698 | return -ENOENT; | |
699 | q = qdisc_leaf(p, clid); | |
700 | } else { /*ingress */ | |
701 | q = dev->qdisc_ingress; | |
702 | } | |
703 | } else { | |
704 | q = dev->qdisc_sleeping; | |
705 | } | |
706 | ||
707 | /* It may be default qdisc, ignore it */ | |
708 | if (q && q->handle == 0) | |
709 | q = NULL; | |
710 | ||
711 | if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { | |
712 | if (tcm->tcm_handle) { | |
713 | if (q && !(n->nlmsg_flags&NLM_F_REPLACE)) | |
714 | return -EEXIST; | |
715 | if (TC_H_MIN(tcm->tcm_handle)) | |
716 | return -EINVAL; | |
717 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
718 | goto create_n_graft; | |
719 | if (n->nlmsg_flags&NLM_F_EXCL) | |
720 | return -EEXIST; | |
721 | if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)) | |
722 | return -EINVAL; | |
723 | if (q == p || | |
724 | (p && check_loop(q, p, 0))) | |
725 | return -ELOOP; | |
726 | atomic_inc(&q->refcnt); | |
727 | goto graft; | |
728 | } else { | |
729 | if (q == NULL) | |
730 | goto create_n_graft; | |
731 | ||
732 | /* This magic test requires explanation. | |
733 | * | |
734 | * We know, that some child q is already | |
735 | * attached to this parent and have choice: | |
736 | * either to change it or to create/graft new one. | |
737 | * | |
738 | * 1. We are allowed to create/graft only | |
739 | * if CREATE and REPLACE flags are set. | |
740 | * | |
741 | * 2. If EXCL is set, requestor wanted to say, | |
742 | * that qdisc tcm_handle is not expected | |
743 | * to exist, so that we choose create/graft too. | |
744 | * | |
745 | * 3. The last case is when no flags are set. | |
746 | * Alas, it is sort of hole in API, we | |
747 | * cannot decide what to do unambiguously. | |
748 | * For now we select create/graft, if | |
749 | * user gave KIND, which does not match existing. | |
750 | */ | |
751 | if ((n->nlmsg_flags&NLM_F_CREATE) && | |
752 | (n->nlmsg_flags&NLM_F_REPLACE) && | |
753 | ((n->nlmsg_flags&NLM_F_EXCL) || | |
754 | (tca[TCA_KIND-1] && | |
755 | rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)))) | |
756 | goto create_n_graft; | |
757 | } | |
758 | } | |
759 | } else { | |
760 | if (!tcm->tcm_handle) | |
761 | return -EINVAL; | |
762 | q = qdisc_lookup(dev, tcm->tcm_handle); | |
763 | } | |
764 | ||
765 | /* Change qdisc parameters */ | |
766 | if (q == NULL) | |
767 | return -ENOENT; | |
768 | if (n->nlmsg_flags&NLM_F_EXCL) | |
769 | return -EEXIST; | |
770 | if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)) | |
771 | return -EINVAL; | |
772 | err = qdisc_change(q, tca); | |
773 | if (err == 0) | |
774 | qdisc_notify(skb, n, clid, NULL, q); | |
775 | return err; | |
776 | ||
777 | create_n_graft: | |
778 | if (!(n->nlmsg_flags&NLM_F_CREATE)) | |
779 | return -ENOENT; | |
780 | if (clid == TC_H_INGRESS) | |
781 | q = qdisc_create(dev, tcm->tcm_parent, tca, &err); | |
10297b99 | 782 | else |
1da177e4 LT |
783 | q = qdisc_create(dev, tcm->tcm_handle, tca, &err); |
784 | if (q == NULL) { | |
785 | if (err == -EAGAIN) | |
786 | goto replay; | |
787 | return err; | |
788 | } | |
789 | ||
790 | graft: | |
791 | if (1) { | |
792 | struct Qdisc *old_q = NULL; | |
793 | err = qdisc_graft(dev, p, clid, q, &old_q); | |
794 | if (err) { | |
795 | if (q) { | |
796 | spin_lock_bh(&dev->queue_lock); | |
797 | qdisc_destroy(q); | |
798 | spin_unlock_bh(&dev->queue_lock); | |
799 | } | |
800 | return err; | |
801 | } | |
802 | qdisc_notify(skb, n, clid, old_q, q); | |
803 | if (old_q) { | |
804 | spin_lock_bh(&dev->queue_lock); | |
805 | qdisc_destroy(old_q); | |
806 | spin_unlock_bh(&dev->queue_lock); | |
807 | } | |
808 | } | |
809 | return 0; | |
810 | } | |
811 | ||
812 | static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, | |
e431b8c0 | 813 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
814 | { |
815 | struct tcmsg *tcm; | |
816 | struct nlmsghdr *nlh; | |
27a884dc | 817 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
818 | struct gnet_dump d; |
819 | ||
e431b8c0 | 820 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
821 | tcm = NLMSG_DATA(nlh); |
822 | tcm->tcm_family = AF_UNSPEC; | |
9ef1d4c7 PM |
823 | tcm->tcm__pad1 = 0; |
824 | tcm->tcm__pad2 = 0; | |
1da177e4 LT |
825 | tcm->tcm_ifindex = q->dev->ifindex; |
826 | tcm->tcm_parent = clid; | |
827 | tcm->tcm_handle = q->handle; | |
828 | tcm->tcm_info = atomic_read(&q->refcnt); | |
829 | RTA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id); | |
830 | if (q->ops->dump && q->ops->dump(q, skb) < 0) | |
831 | goto rtattr_failure; | |
832 | q->qstats.qlen = q->q.qlen; | |
833 | ||
834 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
835 | TCA_XSTATS, q->stats_lock, &d) < 0) | |
836 | goto rtattr_failure; | |
837 | ||
838 | if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) | |
839 | goto rtattr_failure; | |
840 | ||
841 | if (gnet_stats_copy_basic(&d, &q->bstats) < 0 || | |
842 | #ifdef CONFIG_NET_ESTIMATOR | |
843 | gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || | |
844 | #endif | |
845 | gnet_stats_copy_queue(&d, &q->qstats) < 0) | |
846 | goto rtattr_failure; | |
10297b99 | 847 | |
1da177e4 LT |
848 | if (gnet_stats_finish_copy(&d) < 0) |
849 | goto rtattr_failure; | |
10297b99 | 850 | |
27a884dc | 851 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
852 | return skb->len; |
853 | ||
854 | nlmsg_failure: | |
855 | rtattr_failure: | |
dc5fc579 | 856 | nlmsg_trim(skb, b); |
1da177e4 LT |
857 | return -1; |
858 | } | |
859 | ||
860 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
861 | u32 clid, struct Qdisc *old, struct Qdisc *new) | |
862 | { | |
863 | struct sk_buff *skb; | |
864 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
865 | ||
866 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
867 | if (!skb) | |
868 | return -ENOBUFS; | |
869 | ||
870 | if (old && old->handle) { | |
871 | if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0) | |
872 | goto err_out; | |
873 | } | |
874 | if (new) { | |
875 | if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) | |
876 | goto err_out; | |
877 | } | |
878 | ||
879 | if (skb->len) | |
ac6d439d | 880 | return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
881 | |
882 | err_out: | |
883 | kfree_skb(skb); | |
884 | return -EINVAL; | |
885 | } | |
886 | ||
887 | static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) | |
888 | { | |
889 | int idx, q_idx; | |
890 | int s_idx, s_q_idx; | |
891 | struct net_device *dev; | |
892 | struct Qdisc *q; | |
893 | ||
894 | s_idx = cb->args[0]; | |
895 | s_q_idx = q_idx = cb->args[1]; | |
896 | read_lock(&dev_base_lock); | |
897 | for (dev=dev_base, idx=0; dev; dev = dev->next, idx++) { | |
898 | if (idx < s_idx) | |
899 | continue; | |
900 | if (idx > s_idx) | |
901 | s_q_idx = 0; | |
85670cc1 | 902 | read_lock(&qdisc_tree_lock); |
1da177e4 LT |
903 | q_idx = 0; |
904 | list_for_each_entry(q, &dev->qdisc_list, list) { | |
905 | if (q_idx < s_q_idx) { | |
906 | q_idx++; | |
907 | continue; | |
908 | } | |
909 | if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
910 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) { | |
85670cc1 | 911 | read_unlock(&qdisc_tree_lock); |
1da177e4 LT |
912 | goto done; |
913 | } | |
914 | q_idx++; | |
915 | } | |
85670cc1 | 916 | read_unlock(&qdisc_tree_lock); |
1da177e4 LT |
917 | } |
918 | ||
919 | done: | |
920 | read_unlock(&dev_base_lock); | |
921 | ||
922 | cb->args[0] = idx; | |
923 | cb->args[1] = q_idx; | |
924 | ||
925 | return skb->len; | |
926 | } | |
927 | ||
928 | ||
929 | ||
930 | /************************************************ | |
931 | * Traffic classes manipulation. * | |
932 | ************************************************/ | |
933 | ||
934 | ||
935 | ||
936 | static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
937 | { | |
938 | struct tcmsg *tcm = NLMSG_DATA(n); | |
939 | struct rtattr **tca = arg; | |
940 | struct net_device *dev; | |
941 | struct Qdisc *q = NULL; | |
942 | struct Qdisc_class_ops *cops; | |
943 | unsigned long cl = 0; | |
944 | unsigned long new_cl; | |
945 | u32 pid = tcm->tcm_parent; | |
946 | u32 clid = tcm->tcm_handle; | |
947 | u32 qid = TC_H_MAJ(clid); | |
948 | int err; | |
949 | ||
950 | if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
951 | return -ENODEV; | |
952 | ||
953 | /* | |
954 | parent == TC_H_UNSPEC - unspecified parent. | |
955 | parent == TC_H_ROOT - class is root, which has no parent. | |
956 | parent == X:0 - parent is root class. | |
957 | parent == X:Y - parent is a node in hierarchy. | |
958 | parent == 0:Y - parent is X:Y, where X:0 is qdisc. | |
959 | ||
960 | handle == 0:0 - generate handle from kernel pool. | |
961 | handle == 0:Y - class is X:Y, where X:0 is qdisc. | |
962 | handle == X:Y - clear. | |
963 | handle == X:0 - root class. | |
964 | */ | |
965 | ||
966 | /* Step 1. Determine qdisc handle X:0 */ | |
967 | ||
968 | if (pid != TC_H_ROOT) { | |
969 | u32 qid1 = TC_H_MAJ(pid); | |
970 | ||
971 | if (qid && qid1) { | |
972 | /* If both majors are known, they must be identical. */ | |
973 | if (qid != qid1) | |
974 | return -EINVAL; | |
975 | } else if (qid1) { | |
976 | qid = qid1; | |
977 | } else if (qid == 0) | |
978 | qid = dev->qdisc_sleeping->handle; | |
979 | ||
980 | /* Now qid is genuine qdisc handle consistent | |
981 | both with parent and child. | |
982 | ||
983 | TC_H_MAJ(pid) still may be unspecified, complete it now. | |
984 | */ | |
985 | if (pid) | |
986 | pid = TC_H_MAKE(qid, pid); | |
987 | } else { | |
988 | if (qid == 0) | |
989 | qid = dev->qdisc_sleeping->handle; | |
990 | } | |
991 | ||
992 | /* OK. Locate qdisc */ | |
10297b99 | 993 | if ((q = qdisc_lookup(dev, qid)) == NULL) |
1da177e4 LT |
994 | return -ENOENT; |
995 | ||
996 | /* An check that it supports classes */ | |
997 | cops = q->ops->cl_ops; | |
998 | if (cops == NULL) | |
999 | return -EINVAL; | |
1000 | ||
1001 | /* Now try to get class */ | |
1002 | if (clid == 0) { | |
1003 | if (pid == TC_H_ROOT) | |
1004 | clid = qid; | |
1005 | } else | |
1006 | clid = TC_H_MAKE(qid, clid); | |
1007 | ||
1008 | if (clid) | |
1009 | cl = cops->get(q, clid); | |
1010 | ||
1011 | if (cl == 0) { | |
1012 | err = -ENOENT; | |
1013 | if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE)) | |
1014 | goto out; | |
1015 | } else { | |
1016 | switch (n->nlmsg_type) { | |
10297b99 | 1017 | case RTM_NEWTCLASS: |
1da177e4 LT |
1018 | err = -EEXIST; |
1019 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1020 | goto out; | |
1021 | break; | |
1022 | case RTM_DELTCLASS: | |
1023 | err = cops->delete(q, cl); | |
1024 | if (err == 0) | |
1025 | tclass_notify(skb, n, q, cl, RTM_DELTCLASS); | |
1026 | goto out; | |
1027 | case RTM_GETTCLASS: | |
1028 | err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS); | |
1029 | goto out; | |
1030 | default: | |
1031 | err = -EINVAL; | |
1032 | goto out; | |
1033 | } | |
1034 | } | |
1035 | ||
1036 | new_cl = cl; | |
1037 | err = cops->change(q, clid, pid, tca, &new_cl); | |
1038 | if (err == 0) | |
1039 | tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS); | |
1040 | ||
1041 | out: | |
1042 | if (cl) | |
1043 | cops->put(q, cl); | |
1044 | ||
1045 | return err; | |
1046 | } | |
1047 | ||
1048 | ||
1049 | static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, | |
1050 | unsigned long cl, | |
e431b8c0 | 1051 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
1052 | { |
1053 | struct tcmsg *tcm; | |
1054 | struct nlmsghdr *nlh; | |
27a884dc | 1055 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
1056 | struct gnet_dump d; |
1057 | struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; | |
1058 | ||
e431b8c0 | 1059 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
1060 | tcm = NLMSG_DATA(nlh); |
1061 | tcm->tcm_family = AF_UNSPEC; | |
1062 | tcm->tcm_ifindex = q->dev->ifindex; | |
1063 | tcm->tcm_parent = q->handle; | |
1064 | tcm->tcm_handle = q->handle; | |
1065 | tcm->tcm_info = 0; | |
1066 | RTA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id); | |
1067 | if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) | |
1068 | goto rtattr_failure; | |
1069 | ||
1070 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
1071 | TCA_XSTATS, q->stats_lock, &d) < 0) | |
1072 | goto rtattr_failure; | |
1073 | ||
1074 | if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) | |
1075 | goto rtattr_failure; | |
1076 | ||
1077 | if (gnet_stats_finish_copy(&d) < 0) | |
1078 | goto rtattr_failure; | |
1079 | ||
27a884dc | 1080 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
1081 | return skb->len; |
1082 | ||
1083 | nlmsg_failure: | |
1084 | rtattr_failure: | |
dc5fc579 | 1085 | nlmsg_trim(skb, b); |
1da177e4 LT |
1086 | return -1; |
1087 | } | |
1088 | ||
1089 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1090 | struct Qdisc *q, unsigned long cl, int event) | |
1091 | { | |
1092 | struct sk_buff *skb; | |
1093 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1094 | ||
1095 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1096 | if (!skb) | |
1097 | return -ENOBUFS; | |
1098 | ||
1099 | if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) { | |
1100 | kfree_skb(skb); | |
1101 | return -EINVAL; | |
1102 | } | |
1103 | ||
ac6d439d | 1104 | return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
1105 | } |
1106 | ||
1107 | struct qdisc_dump_args | |
1108 | { | |
1109 | struct qdisc_walker w; | |
1110 | struct sk_buff *skb; | |
1111 | struct netlink_callback *cb; | |
1112 | }; | |
1113 | ||
1114 | static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) | |
1115 | { | |
1116 | struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; | |
1117 | ||
1118 | return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid, | |
1119 | a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); | |
1120 | } | |
1121 | ||
1122 | static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) | |
1123 | { | |
1124 | int t; | |
1125 | int s_t; | |
1126 | struct net_device *dev; | |
1127 | struct Qdisc *q; | |
1128 | struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); | |
1129 | struct qdisc_dump_args arg; | |
1130 | ||
1131 | if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) | |
1132 | return 0; | |
1133 | if ((dev = dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
1134 | return 0; | |
1135 | ||
1136 | s_t = cb->args[0]; | |
1137 | t = 0; | |
1138 | ||
85670cc1 | 1139 | read_lock(&qdisc_tree_lock); |
1da177e4 LT |
1140 | list_for_each_entry(q, &dev->qdisc_list, list) { |
1141 | if (t < s_t || !q->ops->cl_ops || | |
1142 | (tcm->tcm_parent && | |
1143 | TC_H_MAJ(tcm->tcm_parent) != q->handle)) { | |
1144 | t++; | |
1145 | continue; | |
1146 | } | |
1147 | if (t > s_t) | |
1148 | memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); | |
1149 | arg.w.fn = qdisc_class_dump; | |
1150 | arg.skb = skb; | |
1151 | arg.cb = cb; | |
1152 | arg.w.stop = 0; | |
1153 | arg.w.skip = cb->args[1]; | |
1154 | arg.w.count = 0; | |
1155 | q->ops->cl_ops->walk(q, &arg.w); | |
1156 | cb->args[1] = arg.w.count; | |
1157 | if (arg.w.stop) | |
1158 | break; | |
1159 | t++; | |
1160 | } | |
85670cc1 | 1161 | read_unlock(&qdisc_tree_lock); |
1da177e4 LT |
1162 | |
1163 | cb->args[0] = t; | |
1164 | ||
1165 | dev_put(dev); | |
1166 | return skb->len; | |
1167 | } | |
1168 | ||
1169 | /* Main classifier routine: scans classifier chain attached | |
1170 | to this qdisc, (optionally) tests for protocol and asks | |
1171 | specific classifiers. | |
1172 | */ | |
1173 | int tc_classify(struct sk_buff *skb, struct tcf_proto *tp, | |
1174 | struct tcf_result *res) | |
1175 | { | |
1176 | int err = 0; | |
66c6f529 | 1177 | __be16 protocol = skb->protocol; |
1da177e4 LT |
1178 | #ifdef CONFIG_NET_CLS_ACT |
1179 | struct tcf_proto *otp = tp; | |
1180 | reclassify: | |
1181 | #endif | |
1182 | protocol = skb->protocol; | |
1183 | ||
1184 | for ( ; tp; tp = tp->next) { | |
1185 | if ((tp->protocol == protocol || | |
b6d9bcb0 | 1186 | tp->protocol == htons(ETH_P_ALL)) && |
1da177e4 LT |
1187 | (err = tp->classify(skb, tp, res)) >= 0) { |
1188 | #ifdef CONFIG_NET_CLS_ACT | |
1189 | if ( TC_ACT_RECLASSIFY == err) { | |
1190 | __u32 verd = (__u32) G_TC_VERD(skb->tc_verd); | |
1191 | tp = otp; | |
1192 | ||
1193 | if (MAX_REC_LOOP < verd++) { | |
1194 | printk("rule prio %d protocol %02x reclassify is buggy packet dropped\n", | |
1195 | tp->prio&0xffff, ntohs(tp->protocol)); | |
1196 | return TC_ACT_SHOT; | |
1197 | } | |
1198 | skb->tc_verd = SET_TC_VERD(skb->tc_verd,verd); | |
1199 | goto reclassify; | |
1200 | } else { | |
10297b99 | 1201 | if (skb->tc_verd) |
1da177e4 LT |
1202 | skb->tc_verd = SET_TC_VERD(skb->tc_verd,0); |
1203 | return err; | |
1204 | } | |
1205 | #else | |
1206 | ||
1207 | return err; | |
1208 | #endif | |
1209 | } | |
1210 | ||
1211 | } | |
1212 | return -1; | |
1213 | } | |
1214 | ||
1da177e4 LT |
1215 | #ifdef CONFIG_PROC_FS |
1216 | static int psched_show(struct seq_file *seq, void *v) | |
1217 | { | |
1218 | seq_printf(seq, "%08x %08x %08x %08x\n", | |
641b9e0e | 1219 | (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1), |
514bca32 PM |
1220 | 1000000, |
1221 | (u32)NSEC_PER_SEC/(u32)ktime_to_ns(KTIME_MONOTONIC_RES)); | |
1da177e4 LT |
1222 | |
1223 | return 0; | |
1224 | } | |
1225 | ||
1226 | static int psched_open(struct inode *inode, struct file *file) | |
1227 | { | |
1228 | return single_open(file, psched_show, PDE(inode)->data); | |
1229 | } | |
1230 | ||
da7071d7 | 1231 | static const struct file_operations psched_fops = { |
1da177e4 LT |
1232 | .owner = THIS_MODULE, |
1233 | .open = psched_open, | |
1234 | .read = seq_read, | |
1235 | .llseek = seq_lseek, | |
1236 | .release = single_release, | |
10297b99 | 1237 | }; |
1da177e4 LT |
1238 | #endif |
1239 | ||
1da177e4 LT |
1240 | static int __init pktsched_init(void) |
1241 | { | |
1242 | struct rtnetlink_link *link_p; | |
1243 | ||
1da177e4 LT |
1244 | link_p = rtnetlink_links[PF_UNSPEC]; |
1245 | ||
1246 | /* Setup rtnetlink links. It is made here to avoid | |
1247 | exporting large number of public symbols. | |
1248 | */ | |
1249 | ||
1250 | if (link_p) { | |
1251 | link_p[RTM_NEWQDISC-RTM_BASE].doit = tc_modify_qdisc; | |
1252 | link_p[RTM_DELQDISC-RTM_BASE].doit = tc_get_qdisc; | |
1253 | link_p[RTM_GETQDISC-RTM_BASE].doit = tc_get_qdisc; | |
1254 | link_p[RTM_GETQDISC-RTM_BASE].dumpit = tc_dump_qdisc; | |
1255 | link_p[RTM_NEWTCLASS-RTM_BASE].doit = tc_ctl_tclass; | |
1256 | link_p[RTM_DELTCLASS-RTM_BASE].doit = tc_ctl_tclass; | |
1257 | link_p[RTM_GETTCLASS-RTM_BASE].doit = tc_ctl_tclass; | |
1258 | link_p[RTM_GETTCLASS-RTM_BASE].dumpit = tc_dump_tclass; | |
1259 | } | |
1260 | ||
1261 | register_qdisc(&pfifo_qdisc_ops); | |
1262 | register_qdisc(&bfifo_qdisc_ops); | |
1263 | proc_net_fops_create("psched", 0, &psched_fops); | |
1264 | ||
1265 | return 0; | |
1266 | } | |
1267 | ||
1268 | subsys_initcall(pktsched_init); | |
1269 | ||
1270 | EXPORT_SYMBOL(qdisc_get_rtab); | |
1271 | EXPORT_SYMBOL(qdisc_put_rtab); | |
1272 | EXPORT_SYMBOL(register_qdisc); | |
1273 | EXPORT_SYMBOL(unregister_qdisc); | |
1274 | EXPORT_SYMBOL(tc_classify); |