]>
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 | 21 | #include <linux/string.h> |
1da177e4 | 22 | #include <linux/errno.h> |
1da177e4 | 23 | #include <linux/skbuff.h> |
1da177e4 LT |
24 | #include <linux/init.h> |
25 | #include <linux/proc_fs.h> | |
26 | #include <linux/seq_file.h> | |
27 | #include <linux/kmod.h> | |
28 | #include <linux/list.h> | |
4179477f | 29 | #include <linux/hrtimer.h> |
1da177e4 | 30 | |
dc5fc579 | 31 | #include <net/netlink.h> |
1da177e4 LT |
32 | #include <net/pkt_sched.h> |
33 | ||
1da177e4 LT |
34 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid, |
35 | struct Qdisc *old, struct Qdisc *new); | |
36 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
37 | struct Qdisc *q, unsigned long cl, int event); | |
38 | ||
39 | /* | |
40 | ||
41 | Short review. | |
42 | ------------- | |
43 | ||
44 | This file consists of two interrelated parts: | |
45 | ||
46 | 1. queueing disciplines manager frontend. | |
47 | 2. traffic classes manager frontend. | |
48 | ||
49 | Generally, queueing discipline ("qdisc") is a black box, | |
50 | which is able to enqueue packets and to dequeue them (when | |
51 | device is ready to send something) in order and at times | |
52 | determined by algorithm hidden in it. | |
53 | ||
54 | qdisc's are divided to two categories: | |
55 | - "queues", which have no internal structure visible from outside. | |
56 | - "schedulers", which split all the packets to "traffic classes", | |
57 | using "packet classifiers" (look at cls_api.c) | |
58 | ||
59 | In turn, classes may have child qdiscs (as rule, queues) | |
60 | attached to them etc. etc. etc. | |
61 | ||
62 | The goal of the routines in this file is to translate | |
63 | information supplied by user in the form of handles | |
64 | to more intelligible for kernel form, to make some sanity | |
65 | checks and part of work, which is common to all qdiscs | |
66 | and to provide rtnetlink notifications. | |
67 | ||
68 | All real intelligent work is done inside qdisc modules. | |
69 | ||
70 | ||
71 | ||
72 | Every discipline has two major routines: enqueue and dequeue. | |
73 | ||
74 | ---dequeue | |
75 | ||
76 | dequeue usually returns a skb to send. It is allowed to return NULL, | |
77 | but it does not mean that queue is empty, it just means that | |
78 | discipline does not want to send anything this time. | |
79 | Queue is really empty if q->q.qlen == 0. | |
80 | For complicated disciplines with multiple queues q->q is not | |
81 | real packet queue, but however q->q.qlen must be valid. | |
82 | ||
83 | ---enqueue | |
84 | ||
85 | enqueue returns 0, if packet was enqueued successfully. | |
86 | If packet (this one or another one) was dropped, it returns | |
87 | not zero error code. | |
88 | NET_XMIT_DROP - this packet dropped | |
89 | Expected action: do not backoff, but wait until queue will clear. | |
90 | NET_XMIT_CN - probably this packet enqueued, but another one dropped. | |
91 | Expected action: backoff or ignore | |
92 | NET_XMIT_POLICED - dropped by police. | |
93 | Expected action: backoff or error to real-time apps. | |
94 | ||
95 | Auxiliary routines: | |
96 | ||
97 | ---requeue | |
98 | ||
99 | requeues once dequeued packet. It is used for non-standard or | |
100 | just buggy devices, which can defer output even if dev->tbusy=0. | |
101 | ||
102 | ---reset | |
103 | ||
104 | returns qdisc to initial state: purge all buffers, clear all | |
105 | timers, counters (except for statistics) etc. | |
106 | ||
107 | ---init | |
108 | ||
109 | initializes newly created qdisc. | |
110 | ||
111 | ---destroy | |
112 | ||
113 | destroys resources allocated by init and during lifetime of qdisc. | |
114 | ||
115 | ---change | |
116 | ||
117 | changes qdisc parameters. | |
118 | */ | |
119 | ||
120 | /* Protects list of registered TC modules. It is pure SMP lock. */ | |
121 | static DEFINE_RWLOCK(qdisc_mod_lock); | |
122 | ||
123 | ||
124 | /************************************************ | |
125 | * Queueing disciplines manipulation. * | |
126 | ************************************************/ | |
127 | ||
128 | ||
129 | /* The list of all installed queueing disciplines. */ | |
130 | ||
131 | static struct Qdisc_ops *qdisc_base; | |
132 | ||
133 | /* Register/uregister queueing discipline */ | |
134 | ||
135 | int register_qdisc(struct Qdisc_ops *qops) | |
136 | { | |
137 | struct Qdisc_ops *q, **qp; | |
138 | int rc = -EEXIST; | |
139 | ||
140 | write_lock(&qdisc_mod_lock); | |
141 | for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) | |
142 | if (!strcmp(qops->id, q->id)) | |
143 | goto out; | |
144 | ||
145 | if (qops->enqueue == NULL) | |
146 | qops->enqueue = noop_qdisc_ops.enqueue; | |
147 | if (qops->requeue == NULL) | |
148 | qops->requeue = noop_qdisc_ops.requeue; | |
149 | if (qops->dequeue == NULL) | |
150 | qops->dequeue = noop_qdisc_ops.dequeue; | |
151 | ||
152 | qops->next = NULL; | |
153 | *qp = qops; | |
154 | rc = 0; | |
155 | out: | |
156 | write_unlock(&qdisc_mod_lock); | |
157 | return rc; | |
158 | } | |
159 | ||
160 | int unregister_qdisc(struct Qdisc_ops *qops) | |
161 | { | |
162 | struct Qdisc_ops *q, **qp; | |
163 | int err = -ENOENT; | |
164 | ||
165 | write_lock(&qdisc_mod_lock); | |
166 | for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next) | |
167 | if (q == qops) | |
168 | break; | |
169 | if (q) { | |
170 | *qp = q->next; | |
171 | q->next = NULL; | |
172 | err = 0; | |
173 | } | |
174 | write_unlock(&qdisc_mod_lock); | |
175 | return err; | |
176 | } | |
177 | ||
178 | /* We know handle. Find qdisc among all qdisc's attached to device | |
179 | (root qdisc, all its children, children of children etc.) | |
180 | */ | |
181 | ||
0463d4ae | 182 | struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) |
1da177e4 LT |
183 | { |
184 | struct Qdisc *q; | |
185 | ||
1da177e4 | 186 | list_for_each_entry(q, &dev->qdisc_list, list) { |
43effa1e | 187 | if (q->handle == handle) |
1da177e4 | 188 | return q; |
1da177e4 | 189 | } |
1da177e4 LT |
190 | return NULL; |
191 | } | |
192 | ||
193 | static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) | |
194 | { | |
195 | unsigned long cl; | |
196 | struct Qdisc *leaf; | |
197 | struct Qdisc_class_ops *cops = p->ops->cl_ops; | |
198 | ||
199 | if (cops == NULL) | |
200 | return NULL; | |
201 | cl = cops->get(p, classid); | |
202 | ||
203 | if (cl == 0) | |
204 | return NULL; | |
205 | leaf = cops->leaf(p, cl); | |
206 | cops->put(p, cl); | |
207 | return leaf; | |
208 | } | |
209 | ||
210 | /* Find queueing discipline by name */ | |
211 | ||
212 | static struct Qdisc_ops *qdisc_lookup_ops(struct rtattr *kind) | |
213 | { | |
214 | struct Qdisc_ops *q = NULL; | |
215 | ||
216 | if (kind) { | |
217 | read_lock(&qdisc_mod_lock); | |
218 | for (q = qdisc_base; q; q = q->next) { | |
219 | if (rtattr_strcmp(kind, q->id) == 0) { | |
220 | if (!try_module_get(q->owner)) | |
221 | q = NULL; | |
222 | break; | |
223 | } | |
224 | } | |
225 | read_unlock(&qdisc_mod_lock); | |
226 | } | |
227 | return q; | |
228 | } | |
229 | ||
230 | static struct qdisc_rate_table *qdisc_rtab_list; | |
231 | ||
232 | struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct rtattr *tab) | |
233 | { | |
234 | struct qdisc_rate_table *rtab; | |
235 | ||
236 | for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { | |
237 | if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) { | |
238 | rtab->refcnt++; | |
239 | return rtab; | |
240 | } | |
241 | } | |
242 | ||
243 | if (tab == NULL || r->rate == 0 || r->cell_log == 0 || RTA_PAYLOAD(tab) != 1024) | |
244 | return NULL; | |
245 | ||
246 | rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); | |
247 | if (rtab) { | |
248 | rtab->rate = *r; | |
249 | rtab->refcnt = 1; | |
250 | memcpy(rtab->data, RTA_DATA(tab), 1024); | |
251 | rtab->next = qdisc_rtab_list; | |
252 | qdisc_rtab_list = rtab; | |
253 | } | |
254 | return rtab; | |
255 | } | |
256 | ||
257 | void qdisc_put_rtab(struct qdisc_rate_table *tab) | |
258 | { | |
259 | struct qdisc_rate_table *rtab, **rtabp; | |
260 | ||
261 | if (!tab || --tab->refcnt) | |
262 | return; | |
263 | ||
264 | for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) { | |
265 | if (rtab == tab) { | |
266 | *rtabp = rtab->next; | |
267 | kfree(rtab); | |
268 | return; | |
269 | } | |
270 | } | |
271 | } | |
272 | ||
4179477f PM |
273 | static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) |
274 | { | |
275 | struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, | |
276 | timer); | |
1936502d | 277 | struct net_device *dev = wd->qdisc->dev; |
4179477f PM |
278 | |
279 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
11274e5a | 280 | smp_wmb(); |
0621ed2e | 281 | netif_schedule(dev); |
1936502d | 282 | |
4179477f PM |
283 | return HRTIMER_NORESTART; |
284 | } | |
285 | ||
286 | void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) | |
287 | { | |
288 | hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
289 | wd->timer.function = qdisc_watchdog; | |
290 | wd->qdisc = qdisc; | |
291 | } | |
292 | EXPORT_SYMBOL(qdisc_watchdog_init); | |
293 | ||
294 | void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires) | |
295 | { | |
296 | ktime_t time; | |
297 | ||
298 | wd->qdisc->flags |= TCQ_F_THROTTLED; | |
299 | time = ktime_set(0, 0); | |
300 | time = ktime_add_ns(time, PSCHED_US2NS(expires)); | |
301 | hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS); | |
302 | } | |
303 | EXPORT_SYMBOL(qdisc_watchdog_schedule); | |
304 | ||
305 | void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) | |
306 | { | |
307 | hrtimer_cancel(&wd->timer); | |
308 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
309 | } | |
310 | EXPORT_SYMBOL(qdisc_watchdog_cancel); | |
1da177e4 LT |
311 | |
312 | /* Allocate an unique handle from space managed by kernel */ | |
313 | ||
314 | static u32 qdisc_alloc_handle(struct net_device *dev) | |
315 | { | |
316 | int i = 0x10000; | |
317 | static u32 autohandle = TC_H_MAKE(0x80000000U, 0); | |
318 | ||
319 | do { | |
320 | autohandle += TC_H_MAKE(0x10000U, 0); | |
321 | if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) | |
322 | autohandle = TC_H_MAKE(0x80000000U, 0); | |
323 | } while (qdisc_lookup(dev, autohandle) && --i > 0); | |
324 | ||
325 | return i>0 ? autohandle : 0; | |
326 | } | |
327 | ||
328 | /* Attach toplevel qdisc to device dev */ | |
329 | ||
330 | static struct Qdisc * | |
331 | dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc) | |
332 | { | |
333 | struct Qdisc *oqdisc; | |
334 | ||
335 | if (dev->flags & IFF_UP) | |
336 | dev_deactivate(dev); | |
337 | ||
338 | qdisc_lock_tree(dev); | |
339 | if (qdisc && qdisc->flags&TCQ_F_INGRESS) { | |
340 | oqdisc = dev->qdisc_ingress; | |
341 | /* Prune old scheduler */ | |
342 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) { | |
343 | /* delete */ | |
344 | qdisc_reset(oqdisc); | |
345 | dev->qdisc_ingress = NULL; | |
346 | } else { /* new */ | |
347 | dev->qdisc_ingress = qdisc; | |
348 | } | |
349 | ||
350 | } else { | |
351 | ||
352 | oqdisc = dev->qdisc_sleeping; | |
353 | ||
354 | /* Prune old scheduler */ | |
355 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
356 | qdisc_reset(oqdisc); | |
357 | ||
358 | /* ... and graft new one */ | |
359 | if (qdisc == NULL) | |
360 | qdisc = &noop_qdisc; | |
361 | dev->qdisc_sleeping = qdisc; | |
362 | dev->qdisc = &noop_qdisc; | |
363 | } | |
364 | ||
365 | qdisc_unlock_tree(dev); | |
366 | ||
367 | if (dev->flags & IFF_UP) | |
368 | dev_activate(dev); | |
369 | ||
370 | return oqdisc; | |
371 | } | |
372 | ||
43effa1e PM |
373 | void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n) |
374 | { | |
375 | struct Qdisc_class_ops *cops; | |
376 | unsigned long cl; | |
377 | u32 parentid; | |
378 | ||
379 | if (n == 0) | |
380 | return; | |
381 | while ((parentid = sch->parent)) { | |
0463d4ae | 382 | sch = qdisc_lookup(sch->dev, TC_H_MAJ(parentid)); |
ffc8fefa PM |
383 | if (sch == NULL) { |
384 | WARN_ON(parentid != TC_H_ROOT); | |
385 | return; | |
386 | } | |
43effa1e PM |
387 | cops = sch->ops->cl_ops; |
388 | if (cops->qlen_notify) { | |
389 | cl = cops->get(sch, parentid); | |
390 | cops->qlen_notify(sch, cl); | |
391 | cops->put(sch, cl); | |
392 | } | |
393 | sch->q.qlen -= n; | |
394 | } | |
395 | } | |
396 | EXPORT_SYMBOL(qdisc_tree_decrease_qlen); | |
1da177e4 LT |
397 | |
398 | /* Graft qdisc "new" to class "classid" of qdisc "parent" or | |
399 | to device "dev". | |
400 | ||
401 | Old qdisc is not destroyed but returned in *old. | |
402 | */ | |
403 | ||
404 | static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, | |
405 | u32 classid, | |
406 | struct Qdisc *new, struct Qdisc **old) | |
407 | { | |
408 | int err = 0; | |
409 | struct Qdisc *q = *old; | |
410 | ||
411 | ||
10297b99 | 412 | if (parent == NULL) { |
1da177e4 LT |
413 | if (q && q->flags&TCQ_F_INGRESS) { |
414 | *old = dev_graft_qdisc(dev, q); | |
415 | } else { | |
416 | *old = dev_graft_qdisc(dev, new); | |
417 | } | |
418 | } else { | |
419 | struct Qdisc_class_ops *cops = parent->ops->cl_ops; | |
420 | ||
421 | err = -EINVAL; | |
422 | ||
423 | if (cops) { | |
424 | unsigned long cl = cops->get(parent, classid); | |
425 | if (cl) { | |
426 | err = cops->graft(parent, cl, new, old); | |
1da177e4 LT |
427 | cops->put(parent, cl); |
428 | } | |
429 | } | |
430 | } | |
431 | return err; | |
432 | } | |
433 | ||
434 | /* | |
435 | Allocate and initialize new qdisc. | |
436 | ||
437 | Parameters are passed via opt. | |
438 | */ | |
439 | ||
440 | static struct Qdisc * | |
ffc8fefa PM |
441 | qdisc_create(struct net_device *dev, u32 parent, u32 handle, |
442 | struct rtattr **tca, int *errp) | |
1da177e4 LT |
443 | { |
444 | int err; | |
445 | struct rtattr *kind = tca[TCA_KIND-1]; | |
1da177e4 LT |
446 | struct Qdisc *sch; |
447 | struct Qdisc_ops *ops; | |
1da177e4 LT |
448 | |
449 | ops = qdisc_lookup_ops(kind); | |
450 | #ifdef CONFIG_KMOD | |
451 | if (ops == NULL && kind != NULL) { | |
452 | char name[IFNAMSIZ]; | |
453 | if (rtattr_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { | |
454 | /* We dropped the RTNL semaphore in order to | |
455 | * perform the module load. So, even if we | |
456 | * succeeded in loading the module we have to | |
457 | * tell the caller to replay the request. We | |
458 | * indicate this using -EAGAIN. | |
459 | * We replay the request because the device may | |
460 | * go away in the mean time. | |
461 | */ | |
462 | rtnl_unlock(); | |
463 | request_module("sch_%s", name); | |
464 | rtnl_lock(); | |
465 | ops = qdisc_lookup_ops(kind); | |
466 | if (ops != NULL) { | |
467 | /* We will try again qdisc_lookup_ops, | |
468 | * so don't keep a reference. | |
469 | */ | |
470 | module_put(ops->owner); | |
471 | err = -EAGAIN; | |
472 | goto err_out; | |
473 | } | |
474 | } | |
475 | } | |
476 | #endif | |
477 | ||
b9e2cc0f | 478 | err = -ENOENT; |
1da177e4 LT |
479 | if (ops == NULL) |
480 | goto err_out; | |
481 | ||
3d54b82f TG |
482 | sch = qdisc_alloc(dev, ops); |
483 | if (IS_ERR(sch)) { | |
484 | err = PTR_ERR(sch); | |
1da177e4 | 485 | goto err_out2; |
3d54b82f | 486 | } |
1da177e4 | 487 | |
ffc8fefa PM |
488 | sch->parent = parent; |
489 | ||
3d54b82f | 490 | if (handle == TC_H_INGRESS) { |
1da177e4 | 491 | sch->flags |= TCQ_F_INGRESS; |
fd44de7c | 492 | sch->stats_lock = &dev->ingress_lock; |
3d54b82f | 493 | handle = TC_H_MAKE(TC_H_INGRESS, 0); |
fd44de7c PM |
494 | } else { |
495 | sch->stats_lock = &dev->queue_lock; | |
496 | if (handle == 0) { | |
497 | handle = qdisc_alloc_handle(dev); | |
498 | err = -ENOMEM; | |
499 | if (handle == 0) | |
500 | goto err_out3; | |
501 | } | |
1da177e4 LT |
502 | } |
503 | ||
3d54b82f | 504 | sch->handle = handle; |
1da177e4 LT |
505 | |
506 | if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS-1])) == 0) { | |
023e09a7 TG |
507 | if (tca[TCA_RATE-1]) { |
508 | err = gen_new_estimator(&sch->bstats, &sch->rate_est, | |
509 | sch->stats_lock, | |
510 | tca[TCA_RATE-1]); | |
511 | if (err) { | |
512 | /* | |
513 | * Any broken qdiscs that would require | |
514 | * a ops->reset() here? The qdisc was never | |
515 | * in action so it shouldn't be necessary. | |
516 | */ | |
517 | if (ops->destroy) | |
518 | ops->destroy(sch); | |
519 | goto err_out3; | |
520 | } | |
521 | } | |
1da177e4 LT |
522 | qdisc_lock_tree(dev); |
523 | list_add_tail(&sch->list, &dev->qdisc_list); | |
524 | qdisc_unlock_tree(dev); | |
525 | ||
1da177e4 LT |
526 | return sch; |
527 | } | |
528 | err_out3: | |
529 | dev_put(dev); | |
3d54b82f | 530 | kfree((char *) sch - sch->padded); |
1da177e4 LT |
531 | err_out2: |
532 | module_put(ops->owner); | |
533 | err_out: | |
534 | *errp = err; | |
1da177e4 LT |
535 | return NULL; |
536 | } | |
537 | ||
538 | static int qdisc_change(struct Qdisc *sch, struct rtattr **tca) | |
539 | { | |
540 | if (tca[TCA_OPTIONS-1]) { | |
541 | int err; | |
542 | ||
543 | if (sch->ops->change == NULL) | |
544 | return -EINVAL; | |
545 | err = sch->ops->change(sch, tca[TCA_OPTIONS-1]); | |
546 | if (err) | |
547 | return err; | |
548 | } | |
1da177e4 LT |
549 | if (tca[TCA_RATE-1]) |
550 | gen_replace_estimator(&sch->bstats, &sch->rate_est, | |
551 | sch->stats_lock, tca[TCA_RATE-1]); | |
1da177e4 LT |
552 | return 0; |
553 | } | |
554 | ||
555 | struct check_loop_arg | |
556 | { | |
557 | struct qdisc_walker w; | |
558 | struct Qdisc *p; | |
559 | int depth; | |
560 | }; | |
561 | ||
562 | static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); | |
563 | ||
564 | static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) | |
565 | { | |
566 | struct check_loop_arg arg; | |
567 | ||
568 | if (q->ops->cl_ops == NULL) | |
569 | return 0; | |
570 | ||
571 | arg.w.stop = arg.w.skip = arg.w.count = 0; | |
572 | arg.w.fn = check_loop_fn; | |
573 | arg.depth = depth; | |
574 | arg.p = p; | |
575 | q->ops->cl_ops->walk(q, &arg.w); | |
576 | return arg.w.stop ? -ELOOP : 0; | |
577 | } | |
578 | ||
579 | static int | |
580 | check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) | |
581 | { | |
582 | struct Qdisc *leaf; | |
583 | struct Qdisc_class_ops *cops = q->ops->cl_ops; | |
584 | struct check_loop_arg *arg = (struct check_loop_arg *)w; | |
585 | ||
586 | leaf = cops->leaf(q, cl); | |
587 | if (leaf) { | |
588 | if (leaf == arg->p || arg->depth > 7) | |
589 | return -ELOOP; | |
590 | return check_loop(leaf, arg->p, arg->depth + 1); | |
591 | } | |
592 | return 0; | |
593 | } | |
594 | ||
595 | /* | |
596 | * Delete/get qdisc. | |
597 | */ | |
598 | ||
599 | static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
600 | { | |
601 | struct tcmsg *tcm = NLMSG_DATA(n); | |
602 | struct rtattr **tca = arg; | |
603 | struct net_device *dev; | |
604 | u32 clid = tcm->tcm_parent; | |
605 | struct Qdisc *q = NULL; | |
606 | struct Qdisc *p = NULL; | |
607 | int err; | |
608 | ||
609 | if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
610 | return -ENODEV; | |
611 | ||
612 | if (clid) { | |
613 | if (clid != TC_H_ROOT) { | |
614 | if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { | |
615 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
616 | return -ENOENT; | |
617 | q = qdisc_leaf(p, clid); | |
618 | } else { /* ingress */ | |
619 | q = dev->qdisc_ingress; | |
10297b99 | 620 | } |
1da177e4 LT |
621 | } else { |
622 | q = dev->qdisc_sleeping; | |
623 | } | |
624 | if (!q) | |
625 | return -ENOENT; | |
626 | ||
627 | if (tcm->tcm_handle && q->handle != tcm->tcm_handle) | |
628 | return -EINVAL; | |
629 | } else { | |
630 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
631 | return -ENOENT; | |
632 | } | |
633 | ||
634 | if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)) | |
635 | return -EINVAL; | |
636 | ||
637 | if (n->nlmsg_type == RTM_DELQDISC) { | |
638 | if (!clid) | |
639 | return -EINVAL; | |
640 | if (q->handle == 0) | |
641 | return -ENOENT; | |
642 | if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0) | |
643 | return err; | |
644 | if (q) { | |
645 | qdisc_notify(skb, n, clid, q, NULL); | |
fd44de7c | 646 | qdisc_lock_tree(dev); |
1da177e4 | 647 | qdisc_destroy(q); |
fd44de7c | 648 | qdisc_unlock_tree(dev); |
1da177e4 LT |
649 | } |
650 | } else { | |
651 | qdisc_notify(skb, n, clid, NULL, q); | |
652 | } | |
653 | return 0; | |
654 | } | |
655 | ||
656 | /* | |
657 | Create/change qdisc. | |
658 | */ | |
659 | ||
660 | static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
661 | { | |
662 | struct tcmsg *tcm; | |
663 | struct rtattr **tca; | |
664 | struct net_device *dev; | |
665 | u32 clid; | |
666 | struct Qdisc *q, *p; | |
667 | int err; | |
668 | ||
669 | replay: | |
670 | /* Reinit, just in case something touches this. */ | |
671 | tcm = NLMSG_DATA(n); | |
672 | tca = arg; | |
673 | clid = tcm->tcm_parent; | |
674 | q = p = NULL; | |
675 | ||
676 | if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
677 | return -ENODEV; | |
678 | ||
679 | if (clid) { | |
680 | if (clid != TC_H_ROOT) { | |
681 | if (clid != TC_H_INGRESS) { | |
682 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
683 | return -ENOENT; | |
684 | q = qdisc_leaf(p, clid); | |
685 | } else { /*ingress */ | |
686 | q = dev->qdisc_ingress; | |
687 | } | |
688 | } else { | |
689 | q = dev->qdisc_sleeping; | |
690 | } | |
691 | ||
692 | /* It may be default qdisc, ignore it */ | |
693 | if (q && q->handle == 0) | |
694 | q = NULL; | |
695 | ||
696 | if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { | |
697 | if (tcm->tcm_handle) { | |
698 | if (q && !(n->nlmsg_flags&NLM_F_REPLACE)) | |
699 | return -EEXIST; | |
700 | if (TC_H_MIN(tcm->tcm_handle)) | |
701 | return -EINVAL; | |
702 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
703 | goto create_n_graft; | |
704 | if (n->nlmsg_flags&NLM_F_EXCL) | |
705 | return -EEXIST; | |
706 | if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)) | |
707 | return -EINVAL; | |
708 | if (q == p || | |
709 | (p && check_loop(q, p, 0))) | |
710 | return -ELOOP; | |
711 | atomic_inc(&q->refcnt); | |
712 | goto graft; | |
713 | } else { | |
714 | if (q == NULL) | |
715 | goto create_n_graft; | |
716 | ||
717 | /* This magic test requires explanation. | |
718 | * | |
719 | * We know, that some child q is already | |
720 | * attached to this parent and have choice: | |
721 | * either to change it or to create/graft new one. | |
722 | * | |
723 | * 1. We are allowed to create/graft only | |
724 | * if CREATE and REPLACE flags are set. | |
725 | * | |
726 | * 2. If EXCL is set, requestor wanted to say, | |
727 | * that qdisc tcm_handle is not expected | |
728 | * to exist, so that we choose create/graft too. | |
729 | * | |
730 | * 3. The last case is when no flags are set. | |
731 | * Alas, it is sort of hole in API, we | |
732 | * cannot decide what to do unambiguously. | |
733 | * For now we select create/graft, if | |
734 | * user gave KIND, which does not match existing. | |
735 | */ | |
736 | if ((n->nlmsg_flags&NLM_F_CREATE) && | |
737 | (n->nlmsg_flags&NLM_F_REPLACE) && | |
738 | ((n->nlmsg_flags&NLM_F_EXCL) || | |
739 | (tca[TCA_KIND-1] && | |
740 | rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)))) | |
741 | goto create_n_graft; | |
742 | } | |
743 | } | |
744 | } else { | |
745 | if (!tcm->tcm_handle) | |
746 | return -EINVAL; | |
747 | q = qdisc_lookup(dev, tcm->tcm_handle); | |
748 | } | |
749 | ||
750 | /* Change qdisc parameters */ | |
751 | if (q == NULL) | |
752 | return -ENOENT; | |
753 | if (n->nlmsg_flags&NLM_F_EXCL) | |
754 | return -EEXIST; | |
755 | if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)) | |
756 | return -EINVAL; | |
757 | err = qdisc_change(q, tca); | |
758 | if (err == 0) | |
759 | qdisc_notify(skb, n, clid, NULL, q); | |
760 | return err; | |
761 | ||
762 | create_n_graft: | |
763 | if (!(n->nlmsg_flags&NLM_F_CREATE)) | |
764 | return -ENOENT; | |
765 | if (clid == TC_H_INGRESS) | |
ffc8fefa PM |
766 | q = qdisc_create(dev, tcm->tcm_parent, tcm->tcm_parent, |
767 | tca, &err); | |
10297b99 | 768 | else |
ffc8fefa PM |
769 | q = qdisc_create(dev, tcm->tcm_parent, tcm->tcm_handle, |
770 | tca, &err); | |
1da177e4 LT |
771 | if (q == NULL) { |
772 | if (err == -EAGAIN) | |
773 | goto replay; | |
774 | return err; | |
775 | } | |
776 | ||
777 | graft: | |
778 | if (1) { | |
779 | struct Qdisc *old_q = NULL; | |
780 | err = qdisc_graft(dev, p, clid, q, &old_q); | |
781 | if (err) { | |
782 | if (q) { | |
fd44de7c | 783 | qdisc_lock_tree(dev); |
1da177e4 | 784 | qdisc_destroy(q); |
fd44de7c | 785 | qdisc_unlock_tree(dev); |
1da177e4 LT |
786 | } |
787 | return err; | |
788 | } | |
789 | qdisc_notify(skb, n, clid, old_q, q); | |
790 | if (old_q) { | |
fd44de7c | 791 | qdisc_lock_tree(dev); |
1da177e4 | 792 | qdisc_destroy(old_q); |
fd44de7c | 793 | qdisc_unlock_tree(dev); |
1da177e4 LT |
794 | } |
795 | } | |
796 | return 0; | |
797 | } | |
798 | ||
799 | static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, | |
e431b8c0 | 800 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
801 | { |
802 | struct tcmsg *tcm; | |
803 | struct nlmsghdr *nlh; | |
27a884dc | 804 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
805 | struct gnet_dump d; |
806 | ||
e431b8c0 | 807 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
808 | tcm = NLMSG_DATA(nlh); |
809 | tcm->tcm_family = AF_UNSPEC; | |
9ef1d4c7 PM |
810 | tcm->tcm__pad1 = 0; |
811 | tcm->tcm__pad2 = 0; | |
1da177e4 LT |
812 | tcm->tcm_ifindex = q->dev->ifindex; |
813 | tcm->tcm_parent = clid; | |
814 | tcm->tcm_handle = q->handle; | |
815 | tcm->tcm_info = atomic_read(&q->refcnt); | |
816 | RTA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id); | |
817 | if (q->ops->dump && q->ops->dump(q, skb) < 0) | |
818 | goto rtattr_failure; | |
819 | q->qstats.qlen = q->q.qlen; | |
820 | ||
821 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
822 | TCA_XSTATS, q->stats_lock, &d) < 0) | |
823 | goto rtattr_failure; | |
824 | ||
825 | if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) | |
826 | goto rtattr_failure; | |
827 | ||
828 | if (gnet_stats_copy_basic(&d, &q->bstats) < 0 || | |
1da177e4 | 829 | gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || |
1da177e4 LT |
830 | gnet_stats_copy_queue(&d, &q->qstats) < 0) |
831 | goto rtattr_failure; | |
10297b99 | 832 | |
1da177e4 LT |
833 | if (gnet_stats_finish_copy(&d) < 0) |
834 | goto rtattr_failure; | |
10297b99 | 835 | |
27a884dc | 836 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
837 | return skb->len; |
838 | ||
839 | nlmsg_failure: | |
840 | rtattr_failure: | |
dc5fc579 | 841 | nlmsg_trim(skb, b); |
1da177e4 LT |
842 | return -1; |
843 | } | |
844 | ||
845 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
846 | u32 clid, struct Qdisc *old, struct Qdisc *new) | |
847 | { | |
848 | struct sk_buff *skb; | |
849 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
850 | ||
851 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
852 | if (!skb) | |
853 | return -ENOBUFS; | |
854 | ||
855 | if (old && old->handle) { | |
856 | if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0) | |
857 | goto err_out; | |
858 | } | |
859 | if (new) { | |
860 | if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) | |
861 | goto err_out; | |
862 | } | |
863 | ||
864 | if (skb->len) | |
ac6d439d | 865 | return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
866 | |
867 | err_out: | |
868 | kfree_skb(skb); | |
869 | return -EINVAL; | |
870 | } | |
871 | ||
872 | static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) | |
873 | { | |
874 | int idx, q_idx; | |
875 | int s_idx, s_q_idx; | |
876 | struct net_device *dev; | |
877 | struct Qdisc *q; | |
878 | ||
879 | s_idx = cb->args[0]; | |
880 | s_q_idx = q_idx = cb->args[1]; | |
881 | read_lock(&dev_base_lock); | |
7562f876 PE |
882 | idx = 0; |
883 | for_each_netdev(dev) { | |
1da177e4 | 884 | if (idx < s_idx) |
7562f876 | 885 | goto cont; |
1da177e4 LT |
886 | if (idx > s_idx) |
887 | s_q_idx = 0; | |
1da177e4 LT |
888 | q_idx = 0; |
889 | list_for_each_entry(q, &dev->qdisc_list, list) { | |
890 | if (q_idx < s_q_idx) { | |
891 | q_idx++; | |
892 | continue; | |
893 | } | |
894 | if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
0463d4ae | 895 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) |
1da177e4 | 896 | goto done; |
1da177e4 LT |
897 | q_idx++; |
898 | } | |
7562f876 PE |
899 | cont: |
900 | idx++; | |
1da177e4 LT |
901 | } |
902 | ||
903 | done: | |
904 | read_unlock(&dev_base_lock); | |
905 | ||
906 | cb->args[0] = idx; | |
907 | cb->args[1] = q_idx; | |
908 | ||
909 | return skb->len; | |
910 | } | |
911 | ||
912 | ||
913 | ||
914 | /************************************************ | |
915 | * Traffic classes manipulation. * | |
916 | ************************************************/ | |
917 | ||
918 | ||
919 | ||
920 | static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
921 | { | |
922 | struct tcmsg *tcm = NLMSG_DATA(n); | |
923 | struct rtattr **tca = arg; | |
924 | struct net_device *dev; | |
925 | struct Qdisc *q = NULL; | |
926 | struct Qdisc_class_ops *cops; | |
927 | unsigned long cl = 0; | |
928 | unsigned long new_cl; | |
929 | u32 pid = tcm->tcm_parent; | |
930 | u32 clid = tcm->tcm_handle; | |
931 | u32 qid = TC_H_MAJ(clid); | |
932 | int err; | |
933 | ||
934 | if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
935 | return -ENODEV; | |
936 | ||
937 | /* | |
938 | parent == TC_H_UNSPEC - unspecified parent. | |
939 | parent == TC_H_ROOT - class is root, which has no parent. | |
940 | parent == X:0 - parent is root class. | |
941 | parent == X:Y - parent is a node in hierarchy. | |
942 | parent == 0:Y - parent is X:Y, where X:0 is qdisc. | |
943 | ||
944 | handle == 0:0 - generate handle from kernel pool. | |
945 | handle == 0:Y - class is X:Y, where X:0 is qdisc. | |
946 | handle == X:Y - clear. | |
947 | handle == X:0 - root class. | |
948 | */ | |
949 | ||
950 | /* Step 1. Determine qdisc handle X:0 */ | |
951 | ||
952 | if (pid != TC_H_ROOT) { | |
953 | u32 qid1 = TC_H_MAJ(pid); | |
954 | ||
955 | if (qid && qid1) { | |
956 | /* If both majors are known, they must be identical. */ | |
957 | if (qid != qid1) | |
958 | return -EINVAL; | |
959 | } else if (qid1) { | |
960 | qid = qid1; | |
961 | } else if (qid == 0) | |
962 | qid = dev->qdisc_sleeping->handle; | |
963 | ||
964 | /* Now qid is genuine qdisc handle consistent | |
965 | both with parent and child. | |
966 | ||
967 | TC_H_MAJ(pid) still may be unspecified, complete it now. | |
968 | */ | |
969 | if (pid) | |
970 | pid = TC_H_MAKE(qid, pid); | |
971 | } else { | |
972 | if (qid == 0) | |
973 | qid = dev->qdisc_sleeping->handle; | |
974 | } | |
975 | ||
976 | /* OK. Locate qdisc */ | |
10297b99 | 977 | if ((q = qdisc_lookup(dev, qid)) == NULL) |
1da177e4 LT |
978 | return -ENOENT; |
979 | ||
980 | /* An check that it supports classes */ | |
981 | cops = q->ops->cl_ops; | |
982 | if (cops == NULL) | |
983 | return -EINVAL; | |
984 | ||
985 | /* Now try to get class */ | |
986 | if (clid == 0) { | |
987 | if (pid == TC_H_ROOT) | |
988 | clid = qid; | |
989 | } else | |
990 | clid = TC_H_MAKE(qid, clid); | |
991 | ||
992 | if (clid) | |
993 | cl = cops->get(q, clid); | |
994 | ||
995 | if (cl == 0) { | |
996 | err = -ENOENT; | |
997 | if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE)) | |
998 | goto out; | |
999 | } else { | |
1000 | switch (n->nlmsg_type) { | |
10297b99 | 1001 | case RTM_NEWTCLASS: |
1da177e4 LT |
1002 | err = -EEXIST; |
1003 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1004 | goto out; | |
1005 | break; | |
1006 | case RTM_DELTCLASS: | |
1007 | err = cops->delete(q, cl); | |
1008 | if (err == 0) | |
1009 | tclass_notify(skb, n, q, cl, RTM_DELTCLASS); | |
1010 | goto out; | |
1011 | case RTM_GETTCLASS: | |
1012 | err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS); | |
1013 | goto out; | |
1014 | default: | |
1015 | err = -EINVAL; | |
1016 | goto out; | |
1017 | } | |
1018 | } | |
1019 | ||
1020 | new_cl = cl; | |
1021 | err = cops->change(q, clid, pid, tca, &new_cl); | |
1022 | if (err == 0) | |
1023 | tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS); | |
1024 | ||
1025 | out: | |
1026 | if (cl) | |
1027 | cops->put(q, cl); | |
1028 | ||
1029 | return err; | |
1030 | } | |
1031 | ||
1032 | ||
1033 | static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, | |
1034 | unsigned long cl, | |
e431b8c0 | 1035 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
1036 | { |
1037 | struct tcmsg *tcm; | |
1038 | struct nlmsghdr *nlh; | |
27a884dc | 1039 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
1040 | struct gnet_dump d; |
1041 | struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; | |
1042 | ||
e431b8c0 | 1043 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
1044 | tcm = NLMSG_DATA(nlh); |
1045 | tcm->tcm_family = AF_UNSPEC; | |
1046 | tcm->tcm_ifindex = q->dev->ifindex; | |
1047 | tcm->tcm_parent = q->handle; | |
1048 | tcm->tcm_handle = q->handle; | |
1049 | tcm->tcm_info = 0; | |
1050 | RTA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id); | |
1051 | if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) | |
1052 | goto rtattr_failure; | |
1053 | ||
1054 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
1055 | TCA_XSTATS, q->stats_lock, &d) < 0) | |
1056 | goto rtattr_failure; | |
1057 | ||
1058 | if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) | |
1059 | goto rtattr_failure; | |
1060 | ||
1061 | if (gnet_stats_finish_copy(&d) < 0) | |
1062 | goto rtattr_failure; | |
1063 | ||
27a884dc | 1064 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
1065 | return skb->len; |
1066 | ||
1067 | nlmsg_failure: | |
1068 | rtattr_failure: | |
dc5fc579 | 1069 | nlmsg_trim(skb, b); |
1da177e4 LT |
1070 | return -1; |
1071 | } | |
1072 | ||
1073 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1074 | struct Qdisc *q, unsigned long cl, int event) | |
1075 | { | |
1076 | struct sk_buff *skb; | |
1077 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1078 | ||
1079 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1080 | if (!skb) | |
1081 | return -ENOBUFS; | |
1082 | ||
1083 | if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) { | |
1084 | kfree_skb(skb); | |
1085 | return -EINVAL; | |
1086 | } | |
1087 | ||
ac6d439d | 1088 | return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
1089 | } |
1090 | ||
1091 | struct qdisc_dump_args | |
1092 | { | |
1093 | struct qdisc_walker w; | |
1094 | struct sk_buff *skb; | |
1095 | struct netlink_callback *cb; | |
1096 | }; | |
1097 | ||
1098 | static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) | |
1099 | { | |
1100 | struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; | |
1101 | ||
1102 | return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid, | |
1103 | a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); | |
1104 | } | |
1105 | ||
1106 | static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) | |
1107 | { | |
1108 | int t; | |
1109 | int s_t; | |
1110 | struct net_device *dev; | |
1111 | struct Qdisc *q; | |
1112 | struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); | |
1113 | struct qdisc_dump_args arg; | |
1114 | ||
1115 | if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) | |
1116 | return 0; | |
1117 | if ((dev = dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
1118 | return 0; | |
1119 | ||
1120 | s_t = cb->args[0]; | |
1121 | t = 0; | |
1122 | ||
1da177e4 LT |
1123 | list_for_each_entry(q, &dev->qdisc_list, list) { |
1124 | if (t < s_t || !q->ops->cl_ops || | |
1125 | (tcm->tcm_parent && | |
1126 | TC_H_MAJ(tcm->tcm_parent) != q->handle)) { | |
1127 | t++; | |
1128 | continue; | |
1129 | } | |
1130 | if (t > s_t) | |
1131 | memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); | |
1132 | arg.w.fn = qdisc_class_dump; | |
1133 | arg.skb = skb; | |
1134 | arg.cb = cb; | |
1135 | arg.w.stop = 0; | |
1136 | arg.w.skip = cb->args[1]; | |
1137 | arg.w.count = 0; | |
1138 | q->ops->cl_ops->walk(q, &arg.w); | |
1139 | cb->args[1] = arg.w.count; | |
1140 | if (arg.w.stop) | |
1141 | break; | |
1142 | t++; | |
1143 | } | |
1da177e4 LT |
1144 | |
1145 | cb->args[0] = t; | |
1146 | ||
1147 | dev_put(dev); | |
1148 | return skb->len; | |
1149 | } | |
1150 | ||
1151 | /* Main classifier routine: scans classifier chain attached | |
1152 | to this qdisc, (optionally) tests for protocol and asks | |
1153 | specific classifiers. | |
1154 | */ | |
73ca4918 PM |
1155 | int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp, |
1156 | struct tcf_result *res) | |
1157 | { | |
1158 | __be16 protocol = skb->protocol; | |
1159 | int err = 0; | |
1160 | ||
1161 | for (; tp; tp = tp->next) { | |
1162 | if ((tp->protocol == protocol || | |
1163 | tp->protocol == htons(ETH_P_ALL)) && | |
1164 | (err = tp->classify(skb, tp, res)) >= 0) { | |
1165 | #ifdef CONFIG_NET_CLS_ACT | |
1166 | if (err != TC_ACT_RECLASSIFY && skb->tc_verd) | |
1167 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0); | |
1168 | #endif | |
1169 | return err; | |
1170 | } | |
1171 | } | |
1172 | return -1; | |
1173 | } | |
1174 | EXPORT_SYMBOL(tc_classify_compat); | |
1175 | ||
1da177e4 | 1176 | int tc_classify(struct sk_buff *skb, struct tcf_proto *tp, |
73ca4918 | 1177 | struct tcf_result *res) |
1da177e4 LT |
1178 | { |
1179 | int err = 0; | |
73ca4918 | 1180 | __be16 protocol; |
1da177e4 LT |
1181 | #ifdef CONFIG_NET_CLS_ACT |
1182 | struct tcf_proto *otp = tp; | |
1183 | reclassify: | |
1184 | #endif | |
1185 | protocol = skb->protocol; | |
1186 | ||
73ca4918 | 1187 | err = tc_classify_compat(skb, tp, res); |
1da177e4 | 1188 | #ifdef CONFIG_NET_CLS_ACT |
73ca4918 PM |
1189 | if (err == TC_ACT_RECLASSIFY) { |
1190 | u32 verd = G_TC_VERD(skb->tc_verd); | |
1191 | tp = otp; | |
1192 | ||
1193 | if (verd++ >= MAX_REC_LOOP) { | |
1194 | printk("rule prio %u protocol %02x reclassify loop, " | |
1195 | "packet dropped\n", | |
1196 | tp->prio&0xffff, ntohs(tp->protocol)); | |
1197 | return TC_ACT_SHOT; | |
1da177e4 | 1198 | } |
73ca4918 PM |
1199 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd); |
1200 | goto reclassify; | |
1da177e4 | 1201 | } |
73ca4918 PM |
1202 | #endif |
1203 | return err; | |
1da177e4 | 1204 | } |
73ca4918 | 1205 | EXPORT_SYMBOL(tc_classify); |
1da177e4 | 1206 | |
a48b5a61 PM |
1207 | void tcf_destroy(struct tcf_proto *tp) |
1208 | { | |
1209 | tp->ops->destroy(tp); | |
1210 | module_put(tp->ops->owner); | |
1211 | kfree(tp); | |
1212 | } | |
1213 | ||
1214 | void tcf_destroy_chain(struct tcf_proto *fl) | |
1215 | { | |
1216 | struct tcf_proto *tp; | |
1217 | ||
1218 | while ((tp = fl) != NULL) { | |
1219 | fl = tp->next; | |
1220 | tcf_destroy(tp); | |
1221 | } | |
1222 | } | |
1223 | EXPORT_SYMBOL(tcf_destroy_chain); | |
1224 | ||
1da177e4 LT |
1225 | #ifdef CONFIG_PROC_FS |
1226 | static int psched_show(struct seq_file *seq, void *v) | |
1227 | { | |
1228 | seq_printf(seq, "%08x %08x %08x %08x\n", | |
641b9e0e | 1229 | (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1), |
514bca32 PM |
1230 | 1000000, |
1231 | (u32)NSEC_PER_SEC/(u32)ktime_to_ns(KTIME_MONOTONIC_RES)); | |
1da177e4 LT |
1232 | |
1233 | return 0; | |
1234 | } | |
1235 | ||
1236 | static int psched_open(struct inode *inode, struct file *file) | |
1237 | { | |
1238 | return single_open(file, psched_show, PDE(inode)->data); | |
1239 | } | |
1240 | ||
da7071d7 | 1241 | static const struct file_operations psched_fops = { |
1da177e4 LT |
1242 | .owner = THIS_MODULE, |
1243 | .open = psched_open, | |
1244 | .read = seq_read, | |
1245 | .llseek = seq_lseek, | |
1246 | .release = single_release, | |
10297b99 | 1247 | }; |
1da177e4 LT |
1248 | #endif |
1249 | ||
1da177e4 LT |
1250 | static int __init pktsched_init(void) |
1251 | { | |
1da177e4 LT |
1252 | register_qdisc(&pfifo_qdisc_ops); |
1253 | register_qdisc(&bfifo_qdisc_ops); | |
1254 | proc_net_fops_create("psched", 0, &psched_fops); | |
1255 | ||
be577ddc TG |
1256 | rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL); |
1257 | rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL); | |
1258 | rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc); | |
1259 | rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL); | |
1260 | rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL); | |
1261 | rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass); | |
1262 | ||
1da177e4 LT |
1263 | return 0; |
1264 | } | |
1265 | ||
1266 | subsys_initcall(pktsched_init); | |
1267 | ||
1268 | EXPORT_SYMBOL(qdisc_get_rtab); | |
1269 | EXPORT_SYMBOL(qdisc_put_rtab); | |
1270 | EXPORT_SYMBOL(register_qdisc); | |
1271 | EXPORT_SYMBOL(unregister_qdisc); |