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> | |
1da177e4 LT |
30 | #include <linux/init.h> |
31 | #include <linux/proc_fs.h> | |
32 | #include <linux/seq_file.h> | |
33 | #include <linux/kmod.h> | |
34 | #include <linux/list.h> | |
35 | #include <linux/bitops.h> | |
4179477f | 36 | #include <linux/hrtimer.h> |
1da177e4 | 37 | |
dc5fc579 | 38 | #include <net/netlink.h> |
1da177e4 LT |
39 | #include <net/sock.h> |
40 | #include <net/pkt_sched.h> | |
41 | ||
42 | #include <asm/processor.h> | |
43 | #include <asm/uaccess.h> | |
44 | #include <asm/system.h> | |
45 | ||
46 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid, | |
47 | struct Qdisc *old, struct Qdisc *new); | |
48 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
49 | struct Qdisc *q, unsigned long cl, int event); | |
50 | ||
51 | /* | |
52 | ||
53 | Short review. | |
54 | ------------- | |
55 | ||
56 | This file consists of two interrelated parts: | |
57 | ||
58 | 1. queueing disciplines manager frontend. | |
59 | 2. traffic classes manager frontend. | |
60 | ||
61 | Generally, queueing discipline ("qdisc") is a black box, | |
62 | which is able to enqueue packets and to dequeue them (when | |
63 | device is ready to send something) in order and at times | |
64 | determined by algorithm hidden in it. | |
65 | ||
66 | qdisc's are divided to two categories: | |
67 | - "queues", which have no internal structure visible from outside. | |
68 | - "schedulers", which split all the packets to "traffic classes", | |
69 | using "packet classifiers" (look at cls_api.c) | |
70 | ||
71 | In turn, classes may have child qdiscs (as rule, queues) | |
72 | attached to them etc. etc. etc. | |
73 | ||
74 | The goal of the routines in this file is to translate | |
75 | information supplied by user in the form of handles | |
76 | to more intelligible for kernel form, to make some sanity | |
77 | checks and part of work, which is common to all qdiscs | |
78 | and to provide rtnetlink notifications. | |
79 | ||
80 | All real intelligent work is done inside qdisc modules. | |
81 | ||
82 | ||
83 | ||
84 | Every discipline has two major routines: enqueue and dequeue. | |
85 | ||
86 | ---dequeue | |
87 | ||
88 | dequeue usually returns a skb to send. It is allowed to return NULL, | |
89 | but it does not mean that queue is empty, it just means that | |
90 | discipline does not want to send anything this time. | |
91 | Queue is really empty if q->q.qlen == 0. | |
92 | For complicated disciplines with multiple queues q->q is not | |
93 | real packet queue, but however q->q.qlen must be valid. | |
94 | ||
95 | ---enqueue | |
96 | ||
97 | enqueue returns 0, if packet was enqueued successfully. | |
98 | If packet (this one or another one) was dropped, it returns | |
99 | not zero error code. | |
100 | NET_XMIT_DROP - this packet dropped | |
101 | Expected action: do not backoff, but wait until queue will clear. | |
102 | NET_XMIT_CN - probably this packet enqueued, but another one dropped. | |
103 | Expected action: backoff or ignore | |
104 | NET_XMIT_POLICED - dropped by police. | |
105 | Expected action: backoff or error to real-time apps. | |
106 | ||
107 | Auxiliary routines: | |
108 | ||
109 | ---requeue | |
110 | ||
111 | requeues once dequeued packet. It is used for non-standard or | |
112 | just buggy devices, which can defer output even if dev->tbusy=0. | |
113 | ||
114 | ---reset | |
115 | ||
116 | returns qdisc to initial state: purge all buffers, clear all | |
117 | timers, counters (except for statistics) etc. | |
118 | ||
119 | ---init | |
120 | ||
121 | initializes newly created qdisc. | |
122 | ||
123 | ---destroy | |
124 | ||
125 | destroys resources allocated by init and during lifetime of qdisc. | |
126 | ||
127 | ---change | |
128 | ||
129 | changes qdisc parameters. | |
130 | */ | |
131 | ||
132 | /* Protects list of registered TC modules. It is pure SMP lock. */ | |
133 | static DEFINE_RWLOCK(qdisc_mod_lock); | |
134 | ||
135 | ||
136 | /************************************************ | |
137 | * Queueing disciplines manipulation. * | |
138 | ************************************************/ | |
139 | ||
140 | ||
141 | /* The list of all installed queueing disciplines. */ | |
142 | ||
143 | static struct Qdisc_ops *qdisc_base; | |
144 | ||
145 | /* Register/uregister queueing discipline */ | |
146 | ||
147 | int register_qdisc(struct Qdisc_ops *qops) | |
148 | { | |
149 | struct Qdisc_ops *q, **qp; | |
150 | int rc = -EEXIST; | |
151 | ||
152 | write_lock(&qdisc_mod_lock); | |
153 | for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) | |
154 | if (!strcmp(qops->id, q->id)) | |
155 | goto out; | |
156 | ||
157 | if (qops->enqueue == NULL) | |
158 | qops->enqueue = noop_qdisc_ops.enqueue; | |
159 | if (qops->requeue == NULL) | |
160 | qops->requeue = noop_qdisc_ops.requeue; | |
161 | if (qops->dequeue == NULL) | |
162 | qops->dequeue = noop_qdisc_ops.dequeue; | |
163 | ||
164 | qops->next = NULL; | |
165 | *qp = qops; | |
166 | rc = 0; | |
167 | out: | |
168 | write_unlock(&qdisc_mod_lock); | |
169 | return rc; | |
170 | } | |
171 | ||
172 | int unregister_qdisc(struct Qdisc_ops *qops) | |
173 | { | |
174 | struct Qdisc_ops *q, **qp; | |
175 | int err = -ENOENT; | |
176 | ||
177 | write_lock(&qdisc_mod_lock); | |
178 | for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next) | |
179 | if (q == qops) | |
180 | break; | |
181 | if (q) { | |
182 | *qp = q->next; | |
183 | q->next = NULL; | |
184 | err = 0; | |
185 | } | |
186 | write_unlock(&qdisc_mod_lock); | |
187 | return err; | |
188 | } | |
189 | ||
190 | /* We know handle. Find qdisc among all qdisc's attached to device | |
191 | (root qdisc, all its children, children of children etc.) | |
192 | */ | |
193 | ||
0463d4ae | 194 | struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) |
1da177e4 LT |
195 | { |
196 | struct Qdisc *q; | |
197 | ||
1da177e4 | 198 | list_for_each_entry(q, &dev->qdisc_list, list) { |
43effa1e | 199 | if (q->handle == handle) |
1da177e4 | 200 | return q; |
1da177e4 | 201 | } |
1da177e4 LT |
202 | return NULL; |
203 | } | |
204 | ||
205 | static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) | |
206 | { | |
207 | unsigned long cl; | |
208 | struct Qdisc *leaf; | |
209 | struct Qdisc_class_ops *cops = p->ops->cl_ops; | |
210 | ||
211 | if (cops == NULL) | |
212 | return NULL; | |
213 | cl = cops->get(p, classid); | |
214 | ||
215 | if (cl == 0) | |
216 | return NULL; | |
217 | leaf = cops->leaf(p, cl); | |
218 | cops->put(p, cl); | |
219 | return leaf; | |
220 | } | |
221 | ||
222 | /* Find queueing discipline by name */ | |
223 | ||
224 | static struct Qdisc_ops *qdisc_lookup_ops(struct rtattr *kind) | |
225 | { | |
226 | struct Qdisc_ops *q = NULL; | |
227 | ||
228 | if (kind) { | |
229 | read_lock(&qdisc_mod_lock); | |
230 | for (q = qdisc_base; q; q = q->next) { | |
231 | if (rtattr_strcmp(kind, q->id) == 0) { | |
232 | if (!try_module_get(q->owner)) | |
233 | q = NULL; | |
234 | break; | |
235 | } | |
236 | } | |
237 | read_unlock(&qdisc_mod_lock); | |
238 | } | |
239 | return q; | |
240 | } | |
241 | ||
242 | static struct qdisc_rate_table *qdisc_rtab_list; | |
243 | ||
244 | struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct rtattr *tab) | |
245 | { | |
246 | struct qdisc_rate_table *rtab; | |
247 | ||
248 | for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { | |
249 | if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) { | |
250 | rtab->refcnt++; | |
251 | return rtab; | |
252 | } | |
253 | } | |
254 | ||
255 | if (tab == NULL || r->rate == 0 || r->cell_log == 0 || RTA_PAYLOAD(tab) != 1024) | |
256 | return NULL; | |
257 | ||
258 | rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); | |
259 | if (rtab) { | |
260 | rtab->rate = *r; | |
261 | rtab->refcnt = 1; | |
262 | memcpy(rtab->data, RTA_DATA(tab), 1024); | |
263 | rtab->next = qdisc_rtab_list; | |
264 | qdisc_rtab_list = rtab; | |
265 | } | |
266 | return rtab; | |
267 | } | |
268 | ||
269 | void qdisc_put_rtab(struct qdisc_rate_table *tab) | |
270 | { | |
271 | struct qdisc_rate_table *rtab, **rtabp; | |
272 | ||
273 | if (!tab || --tab->refcnt) | |
274 | return; | |
275 | ||
276 | for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) { | |
277 | if (rtab == tab) { | |
278 | *rtabp = rtab->next; | |
279 | kfree(rtab); | |
280 | return; | |
281 | } | |
282 | } | |
283 | } | |
284 | ||
4179477f PM |
285 | static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) |
286 | { | |
287 | struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, | |
288 | timer); | |
1936502d | 289 | struct net_device *dev = wd->qdisc->dev; |
4179477f PM |
290 | |
291 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
11274e5a | 292 | smp_wmb(); |
1936502d SH |
293 | if (spin_trylock(&dev->queue_lock)) { |
294 | qdisc_run(dev); | |
295 | spin_unlock(&dev->queue_lock); | |
296 | } else | |
297 | netif_schedule(dev); | |
298 | ||
4179477f PM |
299 | return HRTIMER_NORESTART; |
300 | } | |
301 | ||
302 | void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) | |
303 | { | |
304 | hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
305 | wd->timer.function = qdisc_watchdog; | |
306 | wd->qdisc = qdisc; | |
307 | } | |
308 | EXPORT_SYMBOL(qdisc_watchdog_init); | |
309 | ||
310 | void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires) | |
311 | { | |
312 | ktime_t time; | |
313 | ||
314 | wd->qdisc->flags |= TCQ_F_THROTTLED; | |
315 | time = ktime_set(0, 0); | |
316 | time = ktime_add_ns(time, PSCHED_US2NS(expires)); | |
317 | hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS); | |
318 | } | |
319 | EXPORT_SYMBOL(qdisc_watchdog_schedule); | |
320 | ||
321 | void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) | |
322 | { | |
323 | hrtimer_cancel(&wd->timer); | |
324 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
325 | } | |
326 | EXPORT_SYMBOL(qdisc_watchdog_cancel); | |
1da177e4 LT |
327 | |
328 | /* Allocate an unique handle from space managed by kernel */ | |
329 | ||
330 | static u32 qdisc_alloc_handle(struct net_device *dev) | |
331 | { | |
332 | int i = 0x10000; | |
333 | static u32 autohandle = TC_H_MAKE(0x80000000U, 0); | |
334 | ||
335 | do { | |
336 | autohandle += TC_H_MAKE(0x10000U, 0); | |
337 | if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) | |
338 | autohandle = TC_H_MAKE(0x80000000U, 0); | |
339 | } while (qdisc_lookup(dev, autohandle) && --i > 0); | |
340 | ||
341 | return i>0 ? autohandle : 0; | |
342 | } | |
343 | ||
344 | /* Attach toplevel qdisc to device dev */ | |
345 | ||
346 | static struct Qdisc * | |
347 | dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc) | |
348 | { | |
349 | struct Qdisc *oqdisc; | |
350 | ||
351 | if (dev->flags & IFF_UP) | |
352 | dev_deactivate(dev); | |
353 | ||
354 | qdisc_lock_tree(dev); | |
355 | if (qdisc && qdisc->flags&TCQ_F_INGRESS) { | |
356 | oqdisc = dev->qdisc_ingress; | |
357 | /* Prune old scheduler */ | |
358 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) { | |
359 | /* delete */ | |
360 | qdisc_reset(oqdisc); | |
361 | dev->qdisc_ingress = NULL; | |
362 | } else { /* new */ | |
363 | dev->qdisc_ingress = qdisc; | |
364 | } | |
365 | ||
366 | } else { | |
367 | ||
368 | oqdisc = dev->qdisc_sleeping; | |
369 | ||
370 | /* Prune old scheduler */ | |
371 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
372 | qdisc_reset(oqdisc); | |
373 | ||
374 | /* ... and graft new one */ | |
375 | if (qdisc == NULL) | |
376 | qdisc = &noop_qdisc; | |
377 | dev->qdisc_sleeping = qdisc; | |
378 | dev->qdisc = &noop_qdisc; | |
379 | } | |
380 | ||
381 | qdisc_unlock_tree(dev); | |
382 | ||
383 | if (dev->flags & IFF_UP) | |
384 | dev_activate(dev); | |
385 | ||
386 | return oqdisc; | |
387 | } | |
388 | ||
43effa1e PM |
389 | void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n) |
390 | { | |
391 | struct Qdisc_class_ops *cops; | |
392 | unsigned long cl; | |
393 | u32 parentid; | |
394 | ||
395 | if (n == 0) | |
396 | return; | |
397 | while ((parentid = sch->parent)) { | |
0463d4ae | 398 | sch = qdisc_lookup(sch->dev, TC_H_MAJ(parentid)); |
43effa1e PM |
399 | cops = sch->ops->cl_ops; |
400 | if (cops->qlen_notify) { | |
401 | cl = cops->get(sch, parentid); | |
402 | cops->qlen_notify(sch, cl); | |
403 | cops->put(sch, cl); | |
404 | } | |
405 | sch->q.qlen -= n; | |
406 | } | |
407 | } | |
408 | EXPORT_SYMBOL(qdisc_tree_decrease_qlen); | |
1da177e4 LT |
409 | |
410 | /* Graft qdisc "new" to class "classid" of qdisc "parent" or | |
411 | to device "dev". | |
412 | ||
413 | Old qdisc is not destroyed but returned in *old. | |
414 | */ | |
415 | ||
416 | static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, | |
417 | u32 classid, | |
418 | struct Qdisc *new, struct Qdisc **old) | |
419 | { | |
420 | int err = 0; | |
421 | struct Qdisc *q = *old; | |
422 | ||
423 | ||
10297b99 | 424 | if (parent == NULL) { |
1da177e4 LT |
425 | if (q && q->flags&TCQ_F_INGRESS) { |
426 | *old = dev_graft_qdisc(dev, q); | |
427 | } else { | |
428 | *old = dev_graft_qdisc(dev, new); | |
429 | } | |
430 | } else { | |
431 | struct Qdisc_class_ops *cops = parent->ops->cl_ops; | |
432 | ||
433 | err = -EINVAL; | |
434 | ||
435 | if (cops) { | |
436 | unsigned long cl = cops->get(parent, classid); | |
437 | if (cl) { | |
438 | err = cops->graft(parent, cl, new, old); | |
439 | if (new) | |
440 | new->parent = classid; | |
441 | cops->put(parent, cl); | |
442 | } | |
443 | } | |
444 | } | |
445 | return err; | |
446 | } | |
447 | ||
448 | /* | |
449 | Allocate and initialize new qdisc. | |
450 | ||
451 | Parameters are passed via opt. | |
452 | */ | |
453 | ||
454 | static struct Qdisc * | |
455 | qdisc_create(struct net_device *dev, u32 handle, struct rtattr **tca, int *errp) | |
456 | { | |
457 | int err; | |
458 | struct rtattr *kind = tca[TCA_KIND-1]; | |
1da177e4 LT |
459 | struct Qdisc *sch; |
460 | struct Qdisc_ops *ops; | |
1da177e4 LT |
461 | |
462 | ops = qdisc_lookup_ops(kind); | |
463 | #ifdef CONFIG_KMOD | |
464 | if (ops == NULL && kind != NULL) { | |
465 | char name[IFNAMSIZ]; | |
466 | if (rtattr_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { | |
467 | /* We dropped the RTNL semaphore in order to | |
468 | * perform the module load. So, even if we | |
469 | * succeeded in loading the module we have to | |
470 | * tell the caller to replay the request. We | |
471 | * indicate this using -EAGAIN. | |
472 | * We replay the request because the device may | |
473 | * go away in the mean time. | |
474 | */ | |
475 | rtnl_unlock(); | |
476 | request_module("sch_%s", name); | |
477 | rtnl_lock(); | |
478 | ops = qdisc_lookup_ops(kind); | |
479 | if (ops != NULL) { | |
480 | /* We will try again qdisc_lookup_ops, | |
481 | * so don't keep a reference. | |
482 | */ | |
483 | module_put(ops->owner); | |
484 | err = -EAGAIN; | |
485 | goto err_out; | |
486 | } | |
487 | } | |
488 | } | |
489 | #endif | |
490 | ||
b9e2cc0f | 491 | err = -ENOENT; |
1da177e4 LT |
492 | if (ops == NULL) |
493 | goto err_out; | |
494 | ||
3d54b82f TG |
495 | sch = qdisc_alloc(dev, ops); |
496 | if (IS_ERR(sch)) { | |
497 | err = PTR_ERR(sch); | |
1da177e4 | 498 | goto err_out2; |
3d54b82f | 499 | } |
1da177e4 | 500 | |
3d54b82f | 501 | if (handle == TC_H_INGRESS) { |
1da177e4 | 502 | sch->flags |= TCQ_F_INGRESS; |
3d54b82f TG |
503 | handle = TC_H_MAKE(TC_H_INGRESS, 0); |
504 | } else if (handle == 0) { | |
1da177e4 LT |
505 | handle = qdisc_alloc_handle(dev); |
506 | err = -ENOMEM; | |
507 | if (handle == 0) | |
508 | goto err_out3; | |
509 | } | |
510 | ||
3d54b82f | 511 | sch->handle = handle; |
1da177e4 LT |
512 | |
513 | if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS-1])) == 0) { | |
023e09a7 TG |
514 | #ifdef CONFIG_NET_ESTIMATOR |
515 | if (tca[TCA_RATE-1]) { | |
516 | err = gen_new_estimator(&sch->bstats, &sch->rate_est, | |
517 | sch->stats_lock, | |
518 | tca[TCA_RATE-1]); | |
519 | if (err) { | |
520 | /* | |
521 | * Any broken qdiscs that would require | |
522 | * a ops->reset() here? The qdisc was never | |
523 | * in action so it shouldn't be necessary. | |
524 | */ | |
525 | if (ops->destroy) | |
526 | ops->destroy(sch); | |
527 | goto err_out3; | |
528 | } | |
529 | } | |
530 | #endif | |
1da177e4 LT |
531 | qdisc_lock_tree(dev); |
532 | list_add_tail(&sch->list, &dev->qdisc_list); | |
533 | qdisc_unlock_tree(dev); | |
534 | ||
1da177e4 LT |
535 | return sch; |
536 | } | |
537 | err_out3: | |
538 | dev_put(dev); | |
3d54b82f | 539 | kfree((char *) sch - sch->padded); |
1da177e4 LT |
540 | err_out2: |
541 | module_put(ops->owner); | |
542 | err_out: | |
543 | *errp = err; | |
1da177e4 LT |
544 | return NULL; |
545 | } | |
546 | ||
547 | static int qdisc_change(struct Qdisc *sch, struct rtattr **tca) | |
548 | { | |
549 | if (tca[TCA_OPTIONS-1]) { | |
550 | int err; | |
551 | ||
552 | if (sch->ops->change == NULL) | |
553 | return -EINVAL; | |
554 | err = sch->ops->change(sch, tca[TCA_OPTIONS-1]); | |
555 | if (err) | |
556 | return err; | |
557 | } | |
558 | #ifdef CONFIG_NET_ESTIMATOR | |
559 | if (tca[TCA_RATE-1]) | |
560 | gen_replace_estimator(&sch->bstats, &sch->rate_est, | |
561 | sch->stats_lock, tca[TCA_RATE-1]); | |
562 | #endif | |
563 | return 0; | |
564 | } | |
565 | ||
566 | struct check_loop_arg | |
567 | { | |
568 | struct qdisc_walker w; | |
569 | struct Qdisc *p; | |
570 | int depth; | |
571 | }; | |
572 | ||
573 | static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); | |
574 | ||
575 | static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) | |
576 | { | |
577 | struct check_loop_arg arg; | |
578 | ||
579 | if (q->ops->cl_ops == NULL) | |
580 | return 0; | |
581 | ||
582 | arg.w.stop = arg.w.skip = arg.w.count = 0; | |
583 | arg.w.fn = check_loop_fn; | |
584 | arg.depth = depth; | |
585 | arg.p = p; | |
586 | q->ops->cl_ops->walk(q, &arg.w); | |
587 | return arg.w.stop ? -ELOOP : 0; | |
588 | } | |
589 | ||
590 | static int | |
591 | check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) | |
592 | { | |
593 | struct Qdisc *leaf; | |
594 | struct Qdisc_class_ops *cops = q->ops->cl_ops; | |
595 | struct check_loop_arg *arg = (struct check_loop_arg *)w; | |
596 | ||
597 | leaf = cops->leaf(q, cl); | |
598 | if (leaf) { | |
599 | if (leaf == arg->p || arg->depth > 7) | |
600 | return -ELOOP; | |
601 | return check_loop(leaf, arg->p, arg->depth + 1); | |
602 | } | |
603 | return 0; | |
604 | } | |
605 | ||
606 | /* | |
607 | * Delete/get qdisc. | |
608 | */ | |
609 | ||
610 | static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
611 | { | |
612 | struct tcmsg *tcm = NLMSG_DATA(n); | |
613 | struct rtattr **tca = arg; | |
614 | struct net_device *dev; | |
615 | u32 clid = tcm->tcm_parent; | |
616 | struct Qdisc *q = NULL; | |
617 | struct Qdisc *p = NULL; | |
618 | int err; | |
619 | ||
620 | if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
621 | return -ENODEV; | |
622 | ||
623 | if (clid) { | |
624 | if (clid != TC_H_ROOT) { | |
625 | if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { | |
626 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
627 | return -ENOENT; | |
628 | q = qdisc_leaf(p, clid); | |
629 | } else { /* ingress */ | |
630 | q = dev->qdisc_ingress; | |
10297b99 | 631 | } |
1da177e4 LT |
632 | } else { |
633 | q = dev->qdisc_sleeping; | |
634 | } | |
635 | if (!q) | |
636 | return -ENOENT; | |
637 | ||
638 | if (tcm->tcm_handle && q->handle != tcm->tcm_handle) | |
639 | return -EINVAL; | |
640 | } else { | |
641 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
642 | return -ENOENT; | |
643 | } | |
644 | ||
645 | if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)) | |
646 | return -EINVAL; | |
647 | ||
648 | if (n->nlmsg_type == RTM_DELQDISC) { | |
649 | if (!clid) | |
650 | return -EINVAL; | |
651 | if (q->handle == 0) | |
652 | return -ENOENT; | |
653 | if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0) | |
654 | return err; | |
655 | if (q) { | |
656 | qdisc_notify(skb, n, clid, q, NULL); | |
657 | spin_lock_bh(&dev->queue_lock); | |
658 | qdisc_destroy(q); | |
659 | spin_unlock_bh(&dev->queue_lock); | |
660 | } | |
661 | } else { | |
662 | qdisc_notify(skb, n, clid, NULL, q); | |
663 | } | |
664 | return 0; | |
665 | } | |
666 | ||
667 | /* | |
668 | Create/change qdisc. | |
669 | */ | |
670 | ||
671 | static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
672 | { | |
673 | struct tcmsg *tcm; | |
674 | struct rtattr **tca; | |
675 | struct net_device *dev; | |
676 | u32 clid; | |
677 | struct Qdisc *q, *p; | |
678 | int err; | |
679 | ||
680 | replay: | |
681 | /* Reinit, just in case something touches this. */ | |
682 | tcm = NLMSG_DATA(n); | |
683 | tca = arg; | |
684 | clid = tcm->tcm_parent; | |
685 | q = p = NULL; | |
686 | ||
687 | if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
688 | return -ENODEV; | |
689 | ||
690 | if (clid) { | |
691 | if (clid != TC_H_ROOT) { | |
692 | if (clid != TC_H_INGRESS) { | |
693 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
694 | return -ENOENT; | |
695 | q = qdisc_leaf(p, clid); | |
696 | } else { /*ingress */ | |
697 | q = dev->qdisc_ingress; | |
698 | } | |
699 | } else { | |
700 | q = dev->qdisc_sleeping; | |
701 | } | |
702 | ||
703 | /* It may be default qdisc, ignore it */ | |
704 | if (q && q->handle == 0) | |
705 | q = NULL; | |
706 | ||
707 | if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { | |
708 | if (tcm->tcm_handle) { | |
709 | if (q && !(n->nlmsg_flags&NLM_F_REPLACE)) | |
710 | return -EEXIST; | |
711 | if (TC_H_MIN(tcm->tcm_handle)) | |
712 | return -EINVAL; | |
713 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
714 | goto create_n_graft; | |
715 | if (n->nlmsg_flags&NLM_F_EXCL) | |
716 | return -EEXIST; | |
717 | if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)) | |
718 | return -EINVAL; | |
719 | if (q == p || | |
720 | (p && check_loop(q, p, 0))) | |
721 | return -ELOOP; | |
722 | atomic_inc(&q->refcnt); | |
723 | goto graft; | |
724 | } else { | |
725 | if (q == NULL) | |
726 | goto create_n_graft; | |
727 | ||
728 | /* This magic test requires explanation. | |
729 | * | |
730 | * We know, that some child q is already | |
731 | * attached to this parent and have choice: | |
732 | * either to change it or to create/graft new one. | |
733 | * | |
734 | * 1. We are allowed to create/graft only | |
735 | * if CREATE and REPLACE flags are set. | |
736 | * | |
737 | * 2. If EXCL is set, requestor wanted to say, | |
738 | * that qdisc tcm_handle is not expected | |
739 | * to exist, so that we choose create/graft too. | |
740 | * | |
741 | * 3. The last case is when no flags are set. | |
742 | * Alas, it is sort of hole in API, we | |
743 | * cannot decide what to do unambiguously. | |
744 | * For now we select create/graft, if | |
745 | * user gave KIND, which does not match existing. | |
746 | */ | |
747 | if ((n->nlmsg_flags&NLM_F_CREATE) && | |
748 | (n->nlmsg_flags&NLM_F_REPLACE) && | |
749 | ((n->nlmsg_flags&NLM_F_EXCL) || | |
750 | (tca[TCA_KIND-1] && | |
751 | rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)))) | |
752 | goto create_n_graft; | |
753 | } | |
754 | } | |
755 | } else { | |
756 | if (!tcm->tcm_handle) | |
757 | return -EINVAL; | |
758 | q = qdisc_lookup(dev, tcm->tcm_handle); | |
759 | } | |
760 | ||
761 | /* Change qdisc parameters */ | |
762 | if (q == NULL) | |
763 | return -ENOENT; | |
764 | if (n->nlmsg_flags&NLM_F_EXCL) | |
765 | return -EEXIST; | |
766 | if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id)) | |
767 | return -EINVAL; | |
768 | err = qdisc_change(q, tca); | |
769 | if (err == 0) | |
770 | qdisc_notify(skb, n, clid, NULL, q); | |
771 | return err; | |
772 | ||
773 | create_n_graft: | |
774 | if (!(n->nlmsg_flags&NLM_F_CREATE)) | |
775 | return -ENOENT; | |
776 | if (clid == TC_H_INGRESS) | |
777 | q = qdisc_create(dev, tcm->tcm_parent, tca, &err); | |
10297b99 | 778 | else |
1da177e4 LT |
779 | q = qdisc_create(dev, tcm->tcm_handle, tca, &err); |
780 | if (q == NULL) { | |
781 | if (err == -EAGAIN) | |
782 | goto replay; | |
783 | return err; | |
784 | } | |
785 | ||
786 | graft: | |
787 | if (1) { | |
788 | struct Qdisc *old_q = NULL; | |
789 | err = qdisc_graft(dev, p, clid, q, &old_q); | |
790 | if (err) { | |
791 | if (q) { | |
792 | spin_lock_bh(&dev->queue_lock); | |
793 | qdisc_destroy(q); | |
794 | spin_unlock_bh(&dev->queue_lock); | |
795 | } | |
796 | return err; | |
797 | } | |
798 | qdisc_notify(skb, n, clid, old_q, q); | |
799 | if (old_q) { | |
800 | spin_lock_bh(&dev->queue_lock); | |
801 | qdisc_destroy(old_q); | |
802 | spin_unlock_bh(&dev->queue_lock); | |
803 | } | |
804 | } | |
805 | return 0; | |
806 | } | |
807 | ||
808 | static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, | |
e431b8c0 | 809 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
810 | { |
811 | struct tcmsg *tcm; | |
812 | struct nlmsghdr *nlh; | |
27a884dc | 813 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
814 | struct gnet_dump d; |
815 | ||
e431b8c0 | 816 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
817 | tcm = NLMSG_DATA(nlh); |
818 | tcm->tcm_family = AF_UNSPEC; | |
9ef1d4c7 PM |
819 | tcm->tcm__pad1 = 0; |
820 | tcm->tcm__pad2 = 0; | |
1da177e4 LT |
821 | tcm->tcm_ifindex = q->dev->ifindex; |
822 | tcm->tcm_parent = clid; | |
823 | tcm->tcm_handle = q->handle; | |
824 | tcm->tcm_info = atomic_read(&q->refcnt); | |
825 | RTA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id); | |
826 | if (q->ops->dump && q->ops->dump(q, skb) < 0) | |
827 | goto rtattr_failure; | |
828 | q->qstats.qlen = q->q.qlen; | |
829 | ||
830 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
831 | TCA_XSTATS, q->stats_lock, &d) < 0) | |
832 | goto rtattr_failure; | |
833 | ||
834 | if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) | |
835 | goto rtattr_failure; | |
836 | ||
837 | if (gnet_stats_copy_basic(&d, &q->bstats) < 0 || | |
838 | #ifdef CONFIG_NET_ESTIMATOR | |
839 | gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || | |
840 | #endif | |
841 | gnet_stats_copy_queue(&d, &q->qstats) < 0) | |
842 | goto rtattr_failure; | |
10297b99 | 843 | |
1da177e4 LT |
844 | if (gnet_stats_finish_copy(&d) < 0) |
845 | goto rtattr_failure; | |
10297b99 | 846 | |
27a884dc | 847 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
848 | return skb->len; |
849 | ||
850 | nlmsg_failure: | |
851 | rtattr_failure: | |
dc5fc579 | 852 | nlmsg_trim(skb, b); |
1da177e4 LT |
853 | return -1; |
854 | } | |
855 | ||
856 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
857 | u32 clid, struct Qdisc *old, struct Qdisc *new) | |
858 | { | |
859 | struct sk_buff *skb; | |
860 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
861 | ||
862 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
863 | if (!skb) | |
864 | return -ENOBUFS; | |
865 | ||
866 | if (old && old->handle) { | |
867 | if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0) | |
868 | goto err_out; | |
869 | } | |
870 | if (new) { | |
871 | if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) | |
872 | goto err_out; | |
873 | } | |
874 | ||
875 | if (skb->len) | |
ac6d439d | 876 | return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
877 | |
878 | err_out: | |
879 | kfree_skb(skb); | |
880 | return -EINVAL; | |
881 | } | |
882 | ||
883 | static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) | |
884 | { | |
885 | int idx, q_idx; | |
886 | int s_idx, s_q_idx; | |
887 | struct net_device *dev; | |
888 | struct Qdisc *q; | |
889 | ||
890 | s_idx = cb->args[0]; | |
891 | s_q_idx = q_idx = cb->args[1]; | |
892 | read_lock(&dev_base_lock); | |
893 | for (dev=dev_base, idx=0; dev; dev = dev->next, idx++) { | |
894 | if (idx < s_idx) | |
895 | continue; | |
896 | if (idx > s_idx) | |
897 | s_q_idx = 0; | |
1da177e4 LT |
898 | q_idx = 0; |
899 | list_for_each_entry(q, &dev->qdisc_list, list) { | |
900 | if (q_idx < s_q_idx) { | |
901 | q_idx++; | |
902 | continue; | |
903 | } | |
904 | if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
0463d4ae | 905 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) |
1da177e4 | 906 | goto done; |
1da177e4 LT |
907 | q_idx++; |
908 | } | |
1da177e4 LT |
909 | } |
910 | ||
911 | done: | |
912 | read_unlock(&dev_base_lock); | |
913 | ||
914 | cb->args[0] = idx; | |
915 | cb->args[1] = q_idx; | |
916 | ||
917 | return skb->len; | |
918 | } | |
919 | ||
920 | ||
921 | ||
922 | /************************************************ | |
923 | * Traffic classes manipulation. * | |
924 | ************************************************/ | |
925 | ||
926 | ||
927 | ||
928 | static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
929 | { | |
930 | struct tcmsg *tcm = NLMSG_DATA(n); | |
931 | struct rtattr **tca = arg; | |
932 | struct net_device *dev; | |
933 | struct Qdisc *q = NULL; | |
934 | struct Qdisc_class_ops *cops; | |
935 | unsigned long cl = 0; | |
936 | unsigned long new_cl; | |
937 | u32 pid = tcm->tcm_parent; | |
938 | u32 clid = tcm->tcm_handle; | |
939 | u32 qid = TC_H_MAJ(clid); | |
940 | int err; | |
941 | ||
942 | if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
943 | return -ENODEV; | |
944 | ||
945 | /* | |
946 | parent == TC_H_UNSPEC - unspecified parent. | |
947 | parent == TC_H_ROOT - class is root, which has no parent. | |
948 | parent == X:0 - parent is root class. | |
949 | parent == X:Y - parent is a node in hierarchy. | |
950 | parent == 0:Y - parent is X:Y, where X:0 is qdisc. | |
951 | ||
952 | handle == 0:0 - generate handle from kernel pool. | |
953 | handle == 0:Y - class is X:Y, where X:0 is qdisc. | |
954 | handle == X:Y - clear. | |
955 | handle == X:0 - root class. | |
956 | */ | |
957 | ||
958 | /* Step 1. Determine qdisc handle X:0 */ | |
959 | ||
960 | if (pid != TC_H_ROOT) { | |
961 | u32 qid1 = TC_H_MAJ(pid); | |
962 | ||
963 | if (qid && qid1) { | |
964 | /* If both majors are known, they must be identical. */ | |
965 | if (qid != qid1) | |
966 | return -EINVAL; | |
967 | } else if (qid1) { | |
968 | qid = qid1; | |
969 | } else if (qid == 0) | |
970 | qid = dev->qdisc_sleeping->handle; | |
971 | ||
972 | /* Now qid is genuine qdisc handle consistent | |
973 | both with parent and child. | |
974 | ||
975 | TC_H_MAJ(pid) still may be unspecified, complete it now. | |
976 | */ | |
977 | if (pid) | |
978 | pid = TC_H_MAKE(qid, pid); | |
979 | } else { | |
980 | if (qid == 0) | |
981 | qid = dev->qdisc_sleeping->handle; | |
982 | } | |
983 | ||
984 | /* OK. Locate qdisc */ | |
10297b99 | 985 | if ((q = qdisc_lookup(dev, qid)) == NULL) |
1da177e4 LT |
986 | return -ENOENT; |
987 | ||
988 | /* An check that it supports classes */ | |
989 | cops = q->ops->cl_ops; | |
990 | if (cops == NULL) | |
991 | return -EINVAL; | |
992 | ||
993 | /* Now try to get class */ | |
994 | if (clid == 0) { | |
995 | if (pid == TC_H_ROOT) | |
996 | clid = qid; | |
997 | } else | |
998 | clid = TC_H_MAKE(qid, clid); | |
999 | ||
1000 | if (clid) | |
1001 | cl = cops->get(q, clid); | |
1002 | ||
1003 | if (cl == 0) { | |
1004 | err = -ENOENT; | |
1005 | if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE)) | |
1006 | goto out; | |
1007 | } else { | |
1008 | switch (n->nlmsg_type) { | |
10297b99 | 1009 | case RTM_NEWTCLASS: |
1da177e4 LT |
1010 | err = -EEXIST; |
1011 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1012 | goto out; | |
1013 | break; | |
1014 | case RTM_DELTCLASS: | |
1015 | err = cops->delete(q, cl); | |
1016 | if (err == 0) | |
1017 | tclass_notify(skb, n, q, cl, RTM_DELTCLASS); | |
1018 | goto out; | |
1019 | case RTM_GETTCLASS: | |
1020 | err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS); | |
1021 | goto out; | |
1022 | default: | |
1023 | err = -EINVAL; | |
1024 | goto out; | |
1025 | } | |
1026 | } | |
1027 | ||
1028 | new_cl = cl; | |
1029 | err = cops->change(q, clid, pid, tca, &new_cl); | |
1030 | if (err == 0) | |
1031 | tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS); | |
1032 | ||
1033 | out: | |
1034 | if (cl) | |
1035 | cops->put(q, cl); | |
1036 | ||
1037 | return err; | |
1038 | } | |
1039 | ||
1040 | ||
1041 | static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, | |
1042 | unsigned long cl, | |
e431b8c0 | 1043 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
1044 | { |
1045 | struct tcmsg *tcm; | |
1046 | struct nlmsghdr *nlh; | |
27a884dc | 1047 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
1048 | struct gnet_dump d; |
1049 | struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; | |
1050 | ||
e431b8c0 | 1051 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
1052 | tcm = NLMSG_DATA(nlh); |
1053 | tcm->tcm_family = AF_UNSPEC; | |
1054 | tcm->tcm_ifindex = q->dev->ifindex; | |
1055 | tcm->tcm_parent = q->handle; | |
1056 | tcm->tcm_handle = q->handle; | |
1057 | tcm->tcm_info = 0; | |
1058 | RTA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id); | |
1059 | if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) | |
1060 | goto rtattr_failure; | |
1061 | ||
1062 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
1063 | TCA_XSTATS, q->stats_lock, &d) < 0) | |
1064 | goto rtattr_failure; | |
1065 | ||
1066 | if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) | |
1067 | goto rtattr_failure; | |
1068 | ||
1069 | if (gnet_stats_finish_copy(&d) < 0) | |
1070 | goto rtattr_failure; | |
1071 | ||
27a884dc | 1072 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
1073 | return skb->len; |
1074 | ||
1075 | nlmsg_failure: | |
1076 | rtattr_failure: | |
dc5fc579 | 1077 | nlmsg_trim(skb, b); |
1da177e4 LT |
1078 | return -1; |
1079 | } | |
1080 | ||
1081 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1082 | struct Qdisc *q, unsigned long cl, int event) | |
1083 | { | |
1084 | struct sk_buff *skb; | |
1085 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1086 | ||
1087 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1088 | if (!skb) | |
1089 | return -ENOBUFS; | |
1090 | ||
1091 | if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) { | |
1092 | kfree_skb(skb); | |
1093 | return -EINVAL; | |
1094 | } | |
1095 | ||
ac6d439d | 1096 | return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
1097 | } |
1098 | ||
1099 | struct qdisc_dump_args | |
1100 | { | |
1101 | struct qdisc_walker w; | |
1102 | struct sk_buff *skb; | |
1103 | struct netlink_callback *cb; | |
1104 | }; | |
1105 | ||
1106 | static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) | |
1107 | { | |
1108 | struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; | |
1109 | ||
1110 | return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid, | |
1111 | a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); | |
1112 | } | |
1113 | ||
1114 | static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) | |
1115 | { | |
1116 | int t; | |
1117 | int s_t; | |
1118 | struct net_device *dev; | |
1119 | struct Qdisc *q; | |
1120 | struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); | |
1121 | struct qdisc_dump_args arg; | |
1122 | ||
1123 | if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) | |
1124 | return 0; | |
1125 | if ((dev = dev_get_by_index(tcm->tcm_ifindex)) == NULL) | |
1126 | return 0; | |
1127 | ||
1128 | s_t = cb->args[0]; | |
1129 | t = 0; | |
1130 | ||
1da177e4 LT |
1131 | list_for_each_entry(q, &dev->qdisc_list, list) { |
1132 | if (t < s_t || !q->ops->cl_ops || | |
1133 | (tcm->tcm_parent && | |
1134 | TC_H_MAJ(tcm->tcm_parent) != q->handle)) { | |
1135 | t++; | |
1136 | continue; | |
1137 | } | |
1138 | if (t > s_t) | |
1139 | memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); | |
1140 | arg.w.fn = qdisc_class_dump; | |
1141 | arg.skb = skb; | |
1142 | arg.cb = cb; | |
1143 | arg.w.stop = 0; | |
1144 | arg.w.skip = cb->args[1]; | |
1145 | arg.w.count = 0; | |
1146 | q->ops->cl_ops->walk(q, &arg.w); | |
1147 | cb->args[1] = arg.w.count; | |
1148 | if (arg.w.stop) | |
1149 | break; | |
1150 | t++; | |
1151 | } | |
1da177e4 LT |
1152 | |
1153 | cb->args[0] = t; | |
1154 | ||
1155 | dev_put(dev); | |
1156 | return skb->len; | |
1157 | } | |
1158 | ||
1159 | /* Main classifier routine: scans classifier chain attached | |
1160 | to this qdisc, (optionally) tests for protocol and asks | |
1161 | specific classifiers. | |
1162 | */ | |
1163 | int tc_classify(struct sk_buff *skb, struct tcf_proto *tp, | |
1164 | struct tcf_result *res) | |
1165 | { | |
1166 | int err = 0; | |
66c6f529 | 1167 | __be16 protocol = skb->protocol; |
1da177e4 LT |
1168 | #ifdef CONFIG_NET_CLS_ACT |
1169 | struct tcf_proto *otp = tp; | |
1170 | reclassify: | |
1171 | #endif | |
1172 | protocol = skb->protocol; | |
1173 | ||
1174 | for ( ; tp; tp = tp->next) { | |
1175 | if ((tp->protocol == protocol || | |
b6d9bcb0 | 1176 | tp->protocol == htons(ETH_P_ALL)) && |
1da177e4 LT |
1177 | (err = tp->classify(skb, tp, res)) >= 0) { |
1178 | #ifdef CONFIG_NET_CLS_ACT | |
1179 | if ( TC_ACT_RECLASSIFY == err) { | |
1180 | __u32 verd = (__u32) G_TC_VERD(skb->tc_verd); | |
1181 | tp = otp; | |
1182 | ||
1183 | if (MAX_REC_LOOP < verd++) { | |
1184 | printk("rule prio %d protocol %02x reclassify is buggy packet dropped\n", | |
1185 | tp->prio&0xffff, ntohs(tp->protocol)); | |
1186 | return TC_ACT_SHOT; | |
1187 | } | |
1188 | skb->tc_verd = SET_TC_VERD(skb->tc_verd,verd); | |
1189 | goto reclassify; | |
1190 | } else { | |
10297b99 | 1191 | if (skb->tc_verd) |
1da177e4 LT |
1192 | skb->tc_verd = SET_TC_VERD(skb->tc_verd,0); |
1193 | return err; | |
1194 | } | |
1195 | #else | |
1196 | ||
1197 | return err; | |
1198 | #endif | |
1199 | } | |
1200 | ||
1201 | } | |
1202 | return -1; | |
1203 | } | |
1204 | ||
a48b5a61 PM |
1205 | void tcf_destroy(struct tcf_proto *tp) |
1206 | { | |
1207 | tp->ops->destroy(tp); | |
1208 | module_put(tp->ops->owner); | |
1209 | kfree(tp); | |
1210 | } | |
1211 | ||
1212 | void tcf_destroy_chain(struct tcf_proto *fl) | |
1213 | { | |
1214 | struct tcf_proto *tp; | |
1215 | ||
1216 | while ((tp = fl) != NULL) { | |
1217 | fl = tp->next; | |
1218 | tcf_destroy(tp); | |
1219 | } | |
1220 | } | |
1221 | EXPORT_SYMBOL(tcf_destroy_chain); | |
1222 | ||
1da177e4 LT |
1223 | #ifdef CONFIG_PROC_FS |
1224 | static int psched_show(struct seq_file *seq, void *v) | |
1225 | { | |
1226 | seq_printf(seq, "%08x %08x %08x %08x\n", | |
641b9e0e | 1227 | (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1), |
514bca32 PM |
1228 | 1000000, |
1229 | (u32)NSEC_PER_SEC/(u32)ktime_to_ns(KTIME_MONOTONIC_RES)); | |
1da177e4 LT |
1230 | |
1231 | return 0; | |
1232 | } | |
1233 | ||
1234 | static int psched_open(struct inode *inode, struct file *file) | |
1235 | { | |
1236 | return single_open(file, psched_show, PDE(inode)->data); | |
1237 | } | |
1238 | ||
da7071d7 | 1239 | static const struct file_operations psched_fops = { |
1da177e4 LT |
1240 | .owner = THIS_MODULE, |
1241 | .open = psched_open, | |
1242 | .read = seq_read, | |
1243 | .llseek = seq_lseek, | |
1244 | .release = single_release, | |
10297b99 | 1245 | }; |
1da177e4 LT |
1246 | #endif |
1247 | ||
1da177e4 LT |
1248 | static int __init pktsched_init(void) |
1249 | { | |
1da177e4 LT |
1250 | register_qdisc(&pfifo_qdisc_ops); |
1251 | register_qdisc(&bfifo_qdisc_ops); | |
1252 | proc_net_fops_create("psched", 0, &psched_fops); | |
1253 | ||
be577ddc TG |
1254 | rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL); |
1255 | rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL); | |
1256 | rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc); | |
1257 | rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL); | |
1258 | rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL); | |
1259 | rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass); | |
1260 | ||
1da177e4 LT |
1261 | return 0; |
1262 | } | |
1263 | ||
1264 | subsys_initcall(pktsched_init); | |
1265 | ||
1266 | EXPORT_SYMBOL(qdisc_get_rtab); | |
1267 | EXPORT_SYMBOL(qdisc_put_rtab); | |
1268 | EXPORT_SYMBOL(register_qdisc); | |
1269 | EXPORT_SYMBOL(unregister_qdisc); | |
1270 | EXPORT_SYMBOL(tc_classify); |