]>
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 | |
457c4cbc | 31 | #include <net/net_namespace.h> |
b854272b | 32 | #include <net/sock.h> |
dc5fc579 | 33 | #include <net/netlink.h> |
1da177e4 LT |
34 | #include <net/pkt_sched.h> |
35 | ||
1da177e4 LT |
36 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid, |
37 | struct Qdisc *old, struct Qdisc *new); | |
38 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
39 | struct Qdisc *q, unsigned long cl, int event); | |
40 | ||
41 | /* | |
42 | ||
43 | Short review. | |
44 | ------------- | |
45 | ||
46 | This file consists of two interrelated parts: | |
47 | ||
48 | 1. queueing disciplines manager frontend. | |
49 | 2. traffic classes manager frontend. | |
50 | ||
51 | Generally, queueing discipline ("qdisc") is a black box, | |
52 | which is able to enqueue packets and to dequeue them (when | |
53 | device is ready to send something) in order and at times | |
54 | determined by algorithm hidden in it. | |
55 | ||
56 | qdisc's are divided to two categories: | |
57 | - "queues", which have no internal structure visible from outside. | |
58 | - "schedulers", which split all the packets to "traffic classes", | |
59 | using "packet classifiers" (look at cls_api.c) | |
60 | ||
61 | In turn, classes may have child qdiscs (as rule, queues) | |
62 | attached to them etc. etc. etc. | |
63 | ||
64 | The goal of the routines in this file is to translate | |
65 | information supplied by user in the form of handles | |
66 | to more intelligible for kernel form, to make some sanity | |
67 | checks and part of work, which is common to all qdiscs | |
68 | and to provide rtnetlink notifications. | |
69 | ||
70 | All real intelligent work is done inside qdisc modules. | |
71 | ||
72 | ||
73 | ||
74 | Every discipline has two major routines: enqueue and dequeue. | |
75 | ||
76 | ---dequeue | |
77 | ||
78 | dequeue usually returns a skb to send. It is allowed to return NULL, | |
79 | but it does not mean that queue is empty, it just means that | |
80 | discipline does not want to send anything this time. | |
81 | Queue is really empty if q->q.qlen == 0. | |
82 | For complicated disciplines with multiple queues q->q is not | |
83 | real packet queue, but however q->q.qlen must be valid. | |
84 | ||
85 | ---enqueue | |
86 | ||
87 | enqueue returns 0, if packet was enqueued successfully. | |
88 | If packet (this one or another one) was dropped, it returns | |
89 | not zero error code. | |
90 | NET_XMIT_DROP - this packet dropped | |
91 | Expected action: do not backoff, but wait until queue will clear. | |
92 | NET_XMIT_CN - probably this packet enqueued, but another one dropped. | |
93 | Expected action: backoff or ignore | |
94 | NET_XMIT_POLICED - dropped by police. | |
95 | Expected action: backoff or error to real-time apps. | |
96 | ||
97 | Auxiliary routines: | |
98 | ||
99 | ---requeue | |
100 | ||
101 | requeues once dequeued packet. It is used for non-standard or | |
102 | just buggy devices, which can defer output even if dev->tbusy=0. | |
103 | ||
104 | ---reset | |
105 | ||
106 | returns qdisc to initial state: purge all buffers, clear all | |
107 | timers, counters (except for statistics) etc. | |
108 | ||
109 | ---init | |
110 | ||
111 | initializes newly created qdisc. | |
112 | ||
113 | ---destroy | |
114 | ||
115 | destroys resources allocated by init and during lifetime of qdisc. | |
116 | ||
117 | ---change | |
118 | ||
119 | changes qdisc parameters. | |
120 | */ | |
121 | ||
122 | /* Protects list of registered TC modules. It is pure SMP lock. */ | |
123 | static DEFINE_RWLOCK(qdisc_mod_lock); | |
124 | ||
125 | ||
126 | /************************************************ | |
127 | * Queueing disciplines manipulation. * | |
128 | ************************************************/ | |
129 | ||
130 | ||
131 | /* The list of all installed queueing disciplines. */ | |
132 | ||
133 | static struct Qdisc_ops *qdisc_base; | |
134 | ||
135 | /* Register/uregister queueing discipline */ | |
136 | ||
137 | int register_qdisc(struct Qdisc_ops *qops) | |
138 | { | |
139 | struct Qdisc_ops *q, **qp; | |
140 | int rc = -EEXIST; | |
141 | ||
142 | write_lock(&qdisc_mod_lock); | |
143 | for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) | |
144 | if (!strcmp(qops->id, q->id)) | |
145 | goto out; | |
146 | ||
147 | if (qops->enqueue == NULL) | |
148 | qops->enqueue = noop_qdisc_ops.enqueue; | |
149 | if (qops->requeue == NULL) | |
150 | qops->requeue = noop_qdisc_ops.requeue; | |
151 | if (qops->dequeue == NULL) | |
152 | qops->dequeue = noop_qdisc_ops.dequeue; | |
153 | ||
154 | qops->next = NULL; | |
155 | *qp = qops; | |
156 | rc = 0; | |
157 | out: | |
158 | write_unlock(&qdisc_mod_lock); | |
159 | return rc; | |
160 | } | |
62e3ba1b | 161 | EXPORT_SYMBOL(register_qdisc); |
1da177e4 LT |
162 | |
163 | int unregister_qdisc(struct Qdisc_ops *qops) | |
164 | { | |
165 | struct Qdisc_ops *q, **qp; | |
166 | int err = -ENOENT; | |
167 | ||
168 | write_lock(&qdisc_mod_lock); | |
169 | for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next) | |
170 | if (q == qops) | |
171 | break; | |
172 | if (q) { | |
173 | *qp = q->next; | |
174 | q->next = NULL; | |
175 | err = 0; | |
176 | } | |
177 | write_unlock(&qdisc_mod_lock); | |
178 | return err; | |
179 | } | |
62e3ba1b | 180 | EXPORT_SYMBOL(unregister_qdisc); |
1da177e4 LT |
181 | |
182 | /* We know handle. Find qdisc among all qdisc's attached to device | |
183 | (root qdisc, all its children, children of children etc.) | |
184 | */ | |
185 | ||
0463d4ae | 186 | struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) |
1da177e4 LT |
187 | { |
188 | struct Qdisc *q; | |
189 | ||
1da177e4 | 190 | list_for_each_entry(q, &dev->qdisc_list, list) { |
43effa1e | 191 | if (q->handle == handle) |
1da177e4 | 192 | return q; |
1da177e4 | 193 | } |
1da177e4 LT |
194 | return NULL; |
195 | } | |
196 | ||
197 | static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) | |
198 | { | |
199 | unsigned long cl; | |
200 | struct Qdisc *leaf; | |
20fea08b | 201 | const struct Qdisc_class_ops *cops = p->ops->cl_ops; |
1da177e4 LT |
202 | |
203 | if (cops == NULL) | |
204 | return NULL; | |
205 | cl = cops->get(p, classid); | |
206 | ||
207 | if (cl == 0) | |
208 | return NULL; | |
209 | leaf = cops->leaf(p, cl); | |
210 | cops->put(p, cl); | |
211 | return leaf; | |
212 | } | |
213 | ||
214 | /* Find queueing discipline by name */ | |
215 | ||
1e90474c | 216 | static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) |
1da177e4 LT |
217 | { |
218 | struct Qdisc_ops *q = NULL; | |
219 | ||
220 | if (kind) { | |
221 | read_lock(&qdisc_mod_lock); | |
222 | for (q = qdisc_base; q; q = q->next) { | |
1e90474c | 223 | if (nla_strcmp(kind, q->id) == 0) { |
1da177e4 LT |
224 | if (!try_module_get(q->owner)) |
225 | q = NULL; | |
226 | break; | |
227 | } | |
228 | } | |
229 | read_unlock(&qdisc_mod_lock); | |
230 | } | |
231 | return q; | |
232 | } | |
233 | ||
234 | static struct qdisc_rate_table *qdisc_rtab_list; | |
235 | ||
1e90474c | 236 | struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab) |
1da177e4 LT |
237 | { |
238 | struct qdisc_rate_table *rtab; | |
239 | ||
240 | for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { | |
241 | if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) { | |
242 | rtab->refcnt++; | |
243 | return rtab; | |
244 | } | |
245 | } | |
246 | ||
5feb5e1a PM |
247 | if (tab == NULL || r->rate == 0 || r->cell_log == 0 || |
248 | nla_len(tab) != TC_RTAB_SIZE) | |
1da177e4 LT |
249 | return NULL; |
250 | ||
251 | rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); | |
252 | if (rtab) { | |
253 | rtab->rate = *r; | |
254 | rtab->refcnt = 1; | |
1e90474c | 255 | memcpy(rtab->data, nla_data(tab), 1024); |
1da177e4 LT |
256 | rtab->next = qdisc_rtab_list; |
257 | qdisc_rtab_list = rtab; | |
258 | } | |
259 | return rtab; | |
260 | } | |
62e3ba1b | 261 | EXPORT_SYMBOL(qdisc_get_rtab); |
1da177e4 LT |
262 | |
263 | void qdisc_put_rtab(struct qdisc_rate_table *tab) | |
264 | { | |
265 | struct qdisc_rate_table *rtab, **rtabp; | |
266 | ||
267 | if (!tab || --tab->refcnt) | |
268 | return; | |
269 | ||
270 | for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) { | |
271 | if (rtab == tab) { | |
272 | *rtabp = rtab->next; | |
273 | kfree(rtab); | |
274 | return; | |
275 | } | |
276 | } | |
277 | } | |
62e3ba1b | 278 | EXPORT_SYMBOL(qdisc_put_rtab); |
1da177e4 | 279 | |
4179477f PM |
280 | static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) |
281 | { | |
282 | struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, | |
283 | timer); | |
1936502d | 284 | struct net_device *dev = wd->qdisc->dev; |
4179477f PM |
285 | |
286 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
11274e5a | 287 | smp_wmb(); |
0621ed2e | 288 | netif_schedule(dev); |
1936502d | 289 | |
4179477f PM |
290 | return HRTIMER_NORESTART; |
291 | } | |
292 | ||
293 | void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) | |
294 | { | |
295 | hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
296 | wd->timer.function = qdisc_watchdog; | |
297 | wd->qdisc = qdisc; | |
298 | } | |
299 | EXPORT_SYMBOL(qdisc_watchdog_init); | |
300 | ||
301 | void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires) | |
302 | { | |
303 | ktime_t time; | |
304 | ||
305 | wd->qdisc->flags |= TCQ_F_THROTTLED; | |
306 | time = ktime_set(0, 0); | |
307 | time = ktime_add_ns(time, PSCHED_US2NS(expires)); | |
308 | hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS); | |
309 | } | |
310 | EXPORT_SYMBOL(qdisc_watchdog_schedule); | |
311 | ||
312 | void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) | |
313 | { | |
314 | hrtimer_cancel(&wd->timer); | |
315 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
316 | } | |
317 | EXPORT_SYMBOL(qdisc_watchdog_cancel); | |
1da177e4 | 318 | |
6fe1c7a5 PM |
319 | struct hlist_head *qdisc_class_hash_alloc(unsigned int n) |
320 | { | |
321 | unsigned int size = n * sizeof(struct hlist_head), i; | |
322 | struct hlist_head *h; | |
323 | ||
324 | if (size <= PAGE_SIZE) | |
325 | h = kmalloc(size, GFP_KERNEL); | |
326 | else | |
327 | h = (struct hlist_head *) | |
328 | __get_free_pages(GFP_KERNEL, get_order(size)); | |
329 | ||
330 | if (h != NULL) { | |
331 | for (i = 0; i < n; i++) | |
332 | INIT_HLIST_HEAD(&h[i]); | |
333 | } | |
334 | return h; | |
335 | } | |
336 | ||
337 | static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) | |
338 | { | |
339 | unsigned int size = n * sizeof(struct hlist_head); | |
340 | ||
341 | if (size <= PAGE_SIZE) | |
342 | kfree(h); | |
343 | else | |
344 | free_pages((unsigned long)h, get_order(size)); | |
345 | } | |
346 | ||
347 | void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) | |
348 | { | |
349 | struct Qdisc_class_common *cl; | |
350 | struct hlist_node *n, *next; | |
351 | struct hlist_head *nhash, *ohash; | |
352 | unsigned int nsize, nmask, osize; | |
353 | unsigned int i, h; | |
354 | ||
355 | /* Rehash when load factor exceeds 0.75 */ | |
356 | if (clhash->hashelems * 4 <= clhash->hashsize * 3) | |
357 | return; | |
358 | nsize = clhash->hashsize * 2; | |
359 | nmask = nsize - 1; | |
360 | nhash = qdisc_class_hash_alloc(nsize); | |
361 | if (nhash == NULL) | |
362 | return; | |
363 | ||
364 | ohash = clhash->hash; | |
365 | osize = clhash->hashsize; | |
366 | ||
367 | sch_tree_lock(sch); | |
368 | for (i = 0; i < osize; i++) { | |
369 | hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) { | |
370 | h = qdisc_class_hash(cl->classid, nmask); | |
371 | hlist_add_head(&cl->hnode, &nhash[h]); | |
372 | } | |
373 | } | |
374 | clhash->hash = nhash; | |
375 | clhash->hashsize = nsize; | |
376 | clhash->hashmask = nmask; | |
377 | sch_tree_unlock(sch); | |
378 | ||
379 | qdisc_class_hash_free(ohash, osize); | |
380 | } | |
381 | EXPORT_SYMBOL(qdisc_class_hash_grow); | |
382 | ||
383 | int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) | |
384 | { | |
385 | unsigned int size = 4; | |
386 | ||
387 | clhash->hash = qdisc_class_hash_alloc(size); | |
388 | if (clhash->hash == NULL) | |
389 | return -ENOMEM; | |
390 | clhash->hashsize = size; | |
391 | clhash->hashmask = size - 1; | |
392 | clhash->hashelems = 0; | |
393 | return 0; | |
394 | } | |
395 | EXPORT_SYMBOL(qdisc_class_hash_init); | |
396 | ||
397 | void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) | |
398 | { | |
399 | qdisc_class_hash_free(clhash->hash, clhash->hashsize); | |
400 | } | |
401 | EXPORT_SYMBOL(qdisc_class_hash_destroy); | |
402 | ||
403 | void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, | |
404 | struct Qdisc_class_common *cl) | |
405 | { | |
406 | unsigned int h; | |
407 | ||
408 | INIT_HLIST_NODE(&cl->hnode); | |
409 | h = qdisc_class_hash(cl->classid, clhash->hashmask); | |
410 | hlist_add_head(&cl->hnode, &clhash->hash[h]); | |
411 | clhash->hashelems++; | |
412 | } | |
413 | EXPORT_SYMBOL(qdisc_class_hash_insert); | |
414 | ||
415 | void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, | |
416 | struct Qdisc_class_common *cl) | |
417 | { | |
418 | hlist_del(&cl->hnode); | |
419 | clhash->hashelems--; | |
420 | } | |
421 | EXPORT_SYMBOL(qdisc_class_hash_remove); | |
422 | ||
1da177e4 LT |
423 | /* Allocate an unique handle from space managed by kernel */ |
424 | ||
425 | static u32 qdisc_alloc_handle(struct net_device *dev) | |
426 | { | |
427 | int i = 0x10000; | |
428 | static u32 autohandle = TC_H_MAKE(0x80000000U, 0); | |
429 | ||
430 | do { | |
431 | autohandle += TC_H_MAKE(0x10000U, 0); | |
432 | if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) | |
433 | autohandle = TC_H_MAKE(0x80000000U, 0); | |
434 | } while (qdisc_lookup(dev, autohandle) && --i > 0); | |
435 | ||
436 | return i>0 ? autohandle : 0; | |
437 | } | |
438 | ||
439 | /* Attach toplevel qdisc to device dev */ | |
440 | ||
441 | static struct Qdisc * | |
442 | dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc) | |
443 | { | |
444 | struct Qdisc *oqdisc; | |
445 | ||
446 | if (dev->flags & IFF_UP) | |
447 | dev_deactivate(dev); | |
448 | ||
449 | qdisc_lock_tree(dev); | |
450 | if (qdisc && qdisc->flags&TCQ_F_INGRESS) { | |
451 | oqdisc = dev->qdisc_ingress; | |
452 | /* Prune old scheduler */ | |
453 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) { | |
454 | /* delete */ | |
455 | qdisc_reset(oqdisc); | |
456 | dev->qdisc_ingress = NULL; | |
457 | } else { /* new */ | |
458 | dev->qdisc_ingress = qdisc; | |
459 | } | |
460 | ||
461 | } else { | |
462 | ||
463 | oqdisc = dev->qdisc_sleeping; | |
464 | ||
465 | /* Prune old scheduler */ | |
466 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
467 | qdisc_reset(oqdisc); | |
468 | ||
469 | /* ... and graft new one */ | |
470 | if (qdisc == NULL) | |
471 | qdisc = &noop_qdisc; | |
472 | dev->qdisc_sleeping = qdisc; | |
473 | dev->qdisc = &noop_qdisc; | |
474 | } | |
475 | ||
476 | qdisc_unlock_tree(dev); | |
477 | ||
478 | if (dev->flags & IFF_UP) | |
479 | dev_activate(dev); | |
480 | ||
481 | return oqdisc; | |
482 | } | |
483 | ||
43effa1e PM |
484 | void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n) |
485 | { | |
20fea08b | 486 | const struct Qdisc_class_ops *cops; |
43effa1e PM |
487 | unsigned long cl; |
488 | u32 parentid; | |
489 | ||
490 | if (n == 0) | |
491 | return; | |
492 | while ((parentid = sch->parent)) { | |
066a3b5b JP |
493 | if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) |
494 | return; | |
495 | ||
0463d4ae | 496 | sch = qdisc_lookup(sch->dev, TC_H_MAJ(parentid)); |
ffc8fefa PM |
497 | if (sch == NULL) { |
498 | WARN_ON(parentid != TC_H_ROOT); | |
499 | return; | |
500 | } | |
43effa1e PM |
501 | cops = sch->ops->cl_ops; |
502 | if (cops->qlen_notify) { | |
503 | cl = cops->get(sch, parentid); | |
504 | cops->qlen_notify(sch, cl); | |
505 | cops->put(sch, cl); | |
506 | } | |
507 | sch->q.qlen -= n; | |
508 | } | |
509 | } | |
510 | EXPORT_SYMBOL(qdisc_tree_decrease_qlen); | |
1da177e4 LT |
511 | |
512 | /* Graft qdisc "new" to class "classid" of qdisc "parent" or | |
513 | to device "dev". | |
514 | ||
515 | Old qdisc is not destroyed but returned in *old. | |
516 | */ | |
517 | ||
518 | static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, | |
519 | u32 classid, | |
520 | struct Qdisc *new, struct Qdisc **old) | |
521 | { | |
522 | int err = 0; | |
523 | struct Qdisc *q = *old; | |
524 | ||
525 | ||
10297b99 | 526 | if (parent == NULL) { |
1da177e4 LT |
527 | if (q && q->flags&TCQ_F_INGRESS) { |
528 | *old = dev_graft_qdisc(dev, q); | |
529 | } else { | |
530 | *old = dev_graft_qdisc(dev, new); | |
531 | } | |
532 | } else { | |
20fea08b | 533 | const struct Qdisc_class_ops *cops = parent->ops->cl_ops; |
1da177e4 LT |
534 | |
535 | err = -EINVAL; | |
536 | ||
537 | if (cops) { | |
538 | unsigned long cl = cops->get(parent, classid); | |
539 | if (cl) { | |
540 | err = cops->graft(parent, cl, new, old); | |
1da177e4 LT |
541 | cops->put(parent, cl); |
542 | } | |
543 | } | |
544 | } | |
545 | return err; | |
546 | } | |
547 | ||
548 | /* | |
549 | Allocate and initialize new qdisc. | |
550 | ||
551 | Parameters are passed via opt. | |
552 | */ | |
553 | ||
554 | static struct Qdisc * | |
ffc8fefa | 555 | qdisc_create(struct net_device *dev, u32 parent, u32 handle, |
1e90474c | 556 | struct nlattr **tca, int *errp) |
1da177e4 LT |
557 | { |
558 | int err; | |
1e90474c | 559 | struct nlattr *kind = tca[TCA_KIND]; |
1da177e4 LT |
560 | struct Qdisc *sch; |
561 | struct Qdisc_ops *ops; | |
1da177e4 LT |
562 | |
563 | ops = qdisc_lookup_ops(kind); | |
564 | #ifdef CONFIG_KMOD | |
565 | if (ops == NULL && kind != NULL) { | |
566 | char name[IFNAMSIZ]; | |
1e90474c | 567 | if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { |
1da177e4 LT |
568 | /* We dropped the RTNL semaphore in order to |
569 | * perform the module load. So, even if we | |
570 | * succeeded in loading the module we have to | |
571 | * tell the caller to replay the request. We | |
572 | * indicate this using -EAGAIN. | |
573 | * We replay the request because the device may | |
574 | * go away in the mean time. | |
575 | */ | |
576 | rtnl_unlock(); | |
577 | request_module("sch_%s", name); | |
578 | rtnl_lock(); | |
579 | ops = qdisc_lookup_ops(kind); | |
580 | if (ops != NULL) { | |
581 | /* We will try again qdisc_lookup_ops, | |
582 | * so don't keep a reference. | |
583 | */ | |
584 | module_put(ops->owner); | |
585 | err = -EAGAIN; | |
586 | goto err_out; | |
587 | } | |
588 | } | |
589 | } | |
590 | #endif | |
591 | ||
b9e2cc0f | 592 | err = -ENOENT; |
1da177e4 LT |
593 | if (ops == NULL) |
594 | goto err_out; | |
595 | ||
3d54b82f TG |
596 | sch = qdisc_alloc(dev, ops); |
597 | if (IS_ERR(sch)) { | |
598 | err = PTR_ERR(sch); | |
1da177e4 | 599 | goto err_out2; |
3d54b82f | 600 | } |
1da177e4 | 601 | |
ffc8fefa PM |
602 | sch->parent = parent; |
603 | ||
3d54b82f | 604 | if (handle == TC_H_INGRESS) { |
1da177e4 | 605 | sch->flags |= TCQ_F_INGRESS; |
fd44de7c | 606 | sch->stats_lock = &dev->ingress_lock; |
3d54b82f | 607 | handle = TC_H_MAKE(TC_H_INGRESS, 0); |
fd44de7c PM |
608 | } else { |
609 | sch->stats_lock = &dev->queue_lock; | |
610 | if (handle == 0) { | |
611 | handle = qdisc_alloc_handle(dev); | |
612 | err = -ENOMEM; | |
613 | if (handle == 0) | |
614 | goto err_out3; | |
615 | } | |
1da177e4 LT |
616 | } |
617 | ||
3d54b82f | 618 | sch->handle = handle; |
1da177e4 | 619 | |
1e90474c PM |
620 | if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { |
621 | if (tca[TCA_RATE]) { | |
023e09a7 TG |
622 | err = gen_new_estimator(&sch->bstats, &sch->rate_est, |
623 | sch->stats_lock, | |
1e90474c | 624 | tca[TCA_RATE]); |
023e09a7 TG |
625 | if (err) { |
626 | /* | |
627 | * Any broken qdiscs that would require | |
628 | * a ops->reset() here? The qdisc was never | |
629 | * in action so it shouldn't be necessary. | |
630 | */ | |
631 | if (ops->destroy) | |
632 | ops->destroy(sch); | |
633 | goto err_out3; | |
634 | } | |
635 | } | |
1da177e4 LT |
636 | qdisc_lock_tree(dev); |
637 | list_add_tail(&sch->list, &dev->qdisc_list); | |
638 | qdisc_unlock_tree(dev); | |
639 | ||
1da177e4 LT |
640 | return sch; |
641 | } | |
642 | err_out3: | |
643 | dev_put(dev); | |
3d54b82f | 644 | kfree((char *) sch - sch->padded); |
1da177e4 LT |
645 | err_out2: |
646 | module_put(ops->owner); | |
647 | err_out: | |
648 | *errp = err; | |
1da177e4 LT |
649 | return NULL; |
650 | } | |
651 | ||
1e90474c | 652 | static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) |
1da177e4 | 653 | { |
1e90474c | 654 | if (tca[TCA_OPTIONS]) { |
1da177e4 LT |
655 | int err; |
656 | ||
657 | if (sch->ops->change == NULL) | |
658 | return -EINVAL; | |
1e90474c | 659 | err = sch->ops->change(sch, tca[TCA_OPTIONS]); |
1da177e4 LT |
660 | if (err) |
661 | return err; | |
662 | } | |
1e90474c | 663 | if (tca[TCA_RATE]) |
1da177e4 | 664 | gen_replace_estimator(&sch->bstats, &sch->rate_est, |
1e90474c | 665 | sch->stats_lock, tca[TCA_RATE]); |
1da177e4 LT |
666 | return 0; |
667 | } | |
668 | ||
669 | struct check_loop_arg | |
670 | { | |
671 | struct qdisc_walker w; | |
672 | struct Qdisc *p; | |
673 | int depth; | |
674 | }; | |
675 | ||
676 | static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); | |
677 | ||
678 | static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) | |
679 | { | |
680 | struct check_loop_arg arg; | |
681 | ||
682 | if (q->ops->cl_ops == NULL) | |
683 | return 0; | |
684 | ||
685 | arg.w.stop = arg.w.skip = arg.w.count = 0; | |
686 | arg.w.fn = check_loop_fn; | |
687 | arg.depth = depth; | |
688 | arg.p = p; | |
689 | q->ops->cl_ops->walk(q, &arg.w); | |
690 | return arg.w.stop ? -ELOOP : 0; | |
691 | } | |
692 | ||
693 | static int | |
694 | check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) | |
695 | { | |
696 | struct Qdisc *leaf; | |
20fea08b | 697 | const struct Qdisc_class_ops *cops = q->ops->cl_ops; |
1da177e4 LT |
698 | struct check_loop_arg *arg = (struct check_loop_arg *)w; |
699 | ||
700 | leaf = cops->leaf(q, cl); | |
701 | if (leaf) { | |
702 | if (leaf == arg->p || arg->depth > 7) | |
703 | return -ELOOP; | |
704 | return check_loop(leaf, arg->p, arg->depth + 1); | |
705 | } | |
706 | return 0; | |
707 | } | |
708 | ||
709 | /* | |
710 | * Delete/get qdisc. | |
711 | */ | |
712 | ||
713 | static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
714 | { | |
3b1e0a65 | 715 | struct net *net = sock_net(skb->sk); |
1da177e4 | 716 | struct tcmsg *tcm = NLMSG_DATA(n); |
1e90474c | 717 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
718 | struct net_device *dev; |
719 | u32 clid = tcm->tcm_parent; | |
720 | struct Qdisc *q = NULL; | |
721 | struct Qdisc *p = NULL; | |
722 | int err; | |
723 | ||
b854272b DL |
724 | if (net != &init_net) |
725 | return -EINVAL; | |
726 | ||
881d966b | 727 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
728 | return -ENODEV; |
729 | ||
1e90474c PM |
730 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); |
731 | if (err < 0) | |
732 | return err; | |
733 | ||
1da177e4 LT |
734 | if (clid) { |
735 | if (clid != TC_H_ROOT) { | |
736 | if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { | |
737 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
738 | return -ENOENT; | |
739 | q = qdisc_leaf(p, clid); | |
740 | } else { /* ingress */ | |
741 | q = dev->qdisc_ingress; | |
10297b99 | 742 | } |
1da177e4 LT |
743 | } else { |
744 | q = dev->qdisc_sleeping; | |
745 | } | |
746 | if (!q) | |
747 | return -ENOENT; | |
748 | ||
749 | if (tcm->tcm_handle && q->handle != tcm->tcm_handle) | |
750 | return -EINVAL; | |
751 | } else { | |
752 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
753 | return -ENOENT; | |
754 | } | |
755 | ||
1e90474c | 756 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
757 | return -EINVAL; |
758 | ||
759 | if (n->nlmsg_type == RTM_DELQDISC) { | |
760 | if (!clid) | |
761 | return -EINVAL; | |
762 | if (q->handle == 0) | |
763 | return -ENOENT; | |
764 | if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0) | |
765 | return err; | |
766 | if (q) { | |
767 | qdisc_notify(skb, n, clid, q, NULL); | |
fd44de7c | 768 | qdisc_lock_tree(dev); |
1da177e4 | 769 | qdisc_destroy(q); |
fd44de7c | 770 | qdisc_unlock_tree(dev); |
1da177e4 LT |
771 | } |
772 | } else { | |
773 | qdisc_notify(skb, n, clid, NULL, q); | |
774 | } | |
775 | return 0; | |
776 | } | |
777 | ||
778 | /* | |
779 | Create/change qdisc. | |
780 | */ | |
781 | ||
782 | static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
783 | { | |
3b1e0a65 | 784 | struct net *net = sock_net(skb->sk); |
1da177e4 | 785 | struct tcmsg *tcm; |
1e90474c | 786 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
787 | struct net_device *dev; |
788 | u32 clid; | |
789 | struct Qdisc *q, *p; | |
790 | int err; | |
791 | ||
b854272b DL |
792 | if (net != &init_net) |
793 | return -EINVAL; | |
794 | ||
1da177e4 LT |
795 | replay: |
796 | /* Reinit, just in case something touches this. */ | |
797 | tcm = NLMSG_DATA(n); | |
1da177e4 LT |
798 | clid = tcm->tcm_parent; |
799 | q = p = NULL; | |
800 | ||
881d966b | 801 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
802 | return -ENODEV; |
803 | ||
1e90474c PM |
804 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); |
805 | if (err < 0) | |
806 | return err; | |
807 | ||
1da177e4 LT |
808 | if (clid) { |
809 | if (clid != TC_H_ROOT) { | |
810 | if (clid != TC_H_INGRESS) { | |
811 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
812 | return -ENOENT; | |
813 | q = qdisc_leaf(p, clid); | |
814 | } else { /*ingress */ | |
815 | q = dev->qdisc_ingress; | |
816 | } | |
817 | } else { | |
818 | q = dev->qdisc_sleeping; | |
819 | } | |
820 | ||
821 | /* It may be default qdisc, ignore it */ | |
822 | if (q && q->handle == 0) | |
823 | q = NULL; | |
824 | ||
825 | if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { | |
826 | if (tcm->tcm_handle) { | |
827 | if (q && !(n->nlmsg_flags&NLM_F_REPLACE)) | |
828 | return -EEXIST; | |
829 | if (TC_H_MIN(tcm->tcm_handle)) | |
830 | return -EINVAL; | |
831 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
832 | goto create_n_graft; | |
833 | if (n->nlmsg_flags&NLM_F_EXCL) | |
834 | return -EEXIST; | |
1e90474c | 835 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
836 | return -EINVAL; |
837 | if (q == p || | |
838 | (p && check_loop(q, p, 0))) | |
839 | return -ELOOP; | |
840 | atomic_inc(&q->refcnt); | |
841 | goto graft; | |
842 | } else { | |
843 | if (q == NULL) | |
844 | goto create_n_graft; | |
845 | ||
846 | /* This magic test requires explanation. | |
847 | * | |
848 | * We know, that some child q is already | |
849 | * attached to this parent and have choice: | |
850 | * either to change it or to create/graft new one. | |
851 | * | |
852 | * 1. We are allowed to create/graft only | |
853 | * if CREATE and REPLACE flags are set. | |
854 | * | |
855 | * 2. If EXCL is set, requestor wanted to say, | |
856 | * that qdisc tcm_handle is not expected | |
857 | * to exist, so that we choose create/graft too. | |
858 | * | |
859 | * 3. The last case is when no flags are set. | |
860 | * Alas, it is sort of hole in API, we | |
861 | * cannot decide what to do unambiguously. | |
862 | * For now we select create/graft, if | |
863 | * user gave KIND, which does not match existing. | |
864 | */ | |
865 | if ((n->nlmsg_flags&NLM_F_CREATE) && | |
866 | (n->nlmsg_flags&NLM_F_REPLACE) && | |
867 | ((n->nlmsg_flags&NLM_F_EXCL) || | |
1e90474c PM |
868 | (tca[TCA_KIND] && |
869 | nla_strcmp(tca[TCA_KIND], q->ops->id)))) | |
1da177e4 LT |
870 | goto create_n_graft; |
871 | } | |
872 | } | |
873 | } else { | |
874 | if (!tcm->tcm_handle) | |
875 | return -EINVAL; | |
876 | q = qdisc_lookup(dev, tcm->tcm_handle); | |
877 | } | |
878 | ||
879 | /* Change qdisc parameters */ | |
880 | if (q == NULL) | |
881 | return -ENOENT; | |
882 | if (n->nlmsg_flags&NLM_F_EXCL) | |
883 | return -EEXIST; | |
1e90474c | 884 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
885 | return -EINVAL; |
886 | err = qdisc_change(q, tca); | |
887 | if (err == 0) | |
888 | qdisc_notify(skb, n, clid, NULL, q); | |
889 | return err; | |
890 | ||
891 | create_n_graft: | |
892 | if (!(n->nlmsg_flags&NLM_F_CREATE)) | |
893 | return -ENOENT; | |
894 | if (clid == TC_H_INGRESS) | |
ffc8fefa PM |
895 | q = qdisc_create(dev, tcm->tcm_parent, tcm->tcm_parent, |
896 | tca, &err); | |
10297b99 | 897 | else |
ffc8fefa PM |
898 | q = qdisc_create(dev, tcm->tcm_parent, tcm->tcm_handle, |
899 | tca, &err); | |
1da177e4 LT |
900 | if (q == NULL) { |
901 | if (err == -EAGAIN) | |
902 | goto replay; | |
903 | return err; | |
904 | } | |
905 | ||
906 | graft: | |
907 | if (1) { | |
908 | struct Qdisc *old_q = NULL; | |
909 | err = qdisc_graft(dev, p, clid, q, &old_q); | |
910 | if (err) { | |
911 | if (q) { | |
fd44de7c | 912 | qdisc_lock_tree(dev); |
1da177e4 | 913 | qdisc_destroy(q); |
fd44de7c | 914 | qdisc_unlock_tree(dev); |
1da177e4 LT |
915 | } |
916 | return err; | |
917 | } | |
918 | qdisc_notify(skb, n, clid, old_q, q); | |
919 | if (old_q) { | |
fd44de7c | 920 | qdisc_lock_tree(dev); |
1da177e4 | 921 | qdisc_destroy(old_q); |
fd44de7c | 922 | qdisc_unlock_tree(dev); |
1da177e4 LT |
923 | } |
924 | } | |
925 | return 0; | |
926 | } | |
927 | ||
928 | static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, | |
e431b8c0 | 929 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
930 | { |
931 | struct tcmsg *tcm; | |
932 | struct nlmsghdr *nlh; | |
27a884dc | 933 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
934 | struct gnet_dump d; |
935 | ||
e431b8c0 | 936 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
937 | tcm = NLMSG_DATA(nlh); |
938 | tcm->tcm_family = AF_UNSPEC; | |
9ef1d4c7 PM |
939 | tcm->tcm__pad1 = 0; |
940 | tcm->tcm__pad2 = 0; | |
1da177e4 LT |
941 | tcm->tcm_ifindex = q->dev->ifindex; |
942 | tcm->tcm_parent = clid; | |
943 | tcm->tcm_handle = q->handle; | |
944 | tcm->tcm_info = atomic_read(&q->refcnt); | |
57e1c487 | 945 | NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); |
1da177e4 | 946 | if (q->ops->dump && q->ops->dump(q, skb) < 0) |
1e90474c | 947 | goto nla_put_failure; |
1da177e4 LT |
948 | q->qstats.qlen = q->q.qlen; |
949 | ||
950 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
951 | TCA_XSTATS, q->stats_lock, &d) < 0) | |
1e90474c | 952 | goto nla_put_failure; |
1da177e4 LT |
953 | |
954 | if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) | |
1e90474c | 955 | goto nla_put_failure; |
1da177e4 LT |
956 | |
957 | if (gnet_stats_copy_basic(&d, &q->bstats) < 0 || | |
1da177e4 | 958 | gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || |
1da177e4 | 959 | gnet_stats_copy_queue(&d, &q->qstats) < 0) |
1e90474c | 960 | goto nla_put_failure; |
10297b99 | 961 | |
1da177e4 | 962 | if (gnet_stats_finish_copy(&d) < 0) |
1e90474c | 963 | goto nla_put_failure; |
10297b99 | 964 | |
27a884dc | 965 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
966 | return skb->len; |
967 | ||
968 | nlmsg_failure: | |
1e90474c | 969 | nla_put_failure: |
dc5fc579 | 970 | nlmsg_trim(skb, b); |
1da177e4 LT |
971 | return -1; |
972 | } | |
973 | ||
974 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
975 | u32 clid, struct Qdisc *old, struct Qdisc *new) | |
976 | { | |
977 | struct sk_buff *skb; | |
978 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
979 | ||
980 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
981 | if (!skb) | |
982 | return -ENOBUFS; | |
983 | ||
984 | if (old && old->handle) { | |
985 | if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0) | |
986 | goto err_out; | |
987 | } | |
988 | if (new) { | |
989 | if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) | |
990 | goto err_out; | |
991 | } | |
992 | ||
993 | if (skb->len) | |
97c53cac | 994 | return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
995 | |
996 | err_out: | |
997 | kfree_skb(skb); | |
998 | return -EINVAL; | |
999 | } | |
1000 | ||
1001 | static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) | |
1002 | { | |
3b1e0a65 | 1003 | struct net *net = sock_net(skb->sk); |
1da177e4 LT |
1004 | int idx, q_idx; |
1005 | int s_idx, s_q_idx; | |
1006 | struct net_device *dev; | |
1007 | struct Qdisc *q; | |
1008 | ||
b854272b DL |
1009 | if (net != &init_net) |
1010 | return 0; | |
1011 | ||
1da177e4 LT |
1012 | s_idx = cb->args[0]; |
1013 | s_q_idx = q_idx = cb->args[1]; | |
1014 | read_lock(&dev_base_lock); | |
7562f876 | 1015 | idx = 0; |
881d966b | 1016 | for_each_netdev(&init_net, dev) { |
1da177e4 | 1017 | if (idx < s_idx) |
7562f876 | 1018 | goto cont; |
1da177e4 LT |
1019 | if (idx > s_idx) |
1020 | s_q_idx = 0; | |
1da177e4 LT |
1021 | q_idx = 0; |
1022 | list_for_each_entry(q, &dev->qdisc_list, list) { | |
1023 | if (q_idx < s_q_idx) { | |
1024 | q_idx++; | |
1025 | continue; | |
1026 | } | |
1027 | if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
0463d4ae | 1028 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) |
1da177e4 | 1029 | goto done; |
1da177e4 LT |
1030 | q_idx++; |
1031 | } | |
7562f876 PE |
1032 | cont: |
1033 | idx++; | |
1da177e4 LT |
1034 | } |
1035 | ||
1036 | done: | |
1037 | read_unlock(&dev_base_lock); | |
1038 | ||
1039 | cb->args[0] = idx; | |
1040 | cb->args[1] = q_idx; | |
1041 | ||
1042 | return skb->len; | |
1043 | } | |
1044 | ||
1045 | ||
1046 | ||
1047 | /************************************************ | |
1048 | * Traffic classes manipulation. * | |
1049 | ************************************************/ | |
1050 | ||
1051 | ||
1052 | ||
1053 | static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
1054 | { | |
3b1e0a65 | 1055 | struct net *net = sock_net(skb->sk); |
1da177e4 | 1056 | struct tcmsg *tcm = NLMSG_DATA(n); |
1e90474c | 1057 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
1058 | struct net_device *dev; |
1059 | struct Qdisc *q = NULL; | |
20fea08b | 1060 | const struct Qdisc_class_ops *cops; |
1da177e4 LT |
1061 | unsigned long cl = 0; |
1062 | unsigned long new_cl; | |
1063 | u32 pid = tcm->tcm_parent; | |
1064 | u32 clid = tcm->tcm_handle; | |
1065 | u32 qid = TC_H_MAJ(clid); | |
1066 | int err; | |
1067 | ||
b854272b DL |
1068 | if (net != &init_net) |
1069 | return -EINVAL; | |
1070 | ||
881d966b | 1071 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
1072 | return -ENODEV; |
1073 | ||
1e90474c PM |
1074 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); |
1075 | if (err < 0) | |
1076 | return err; | |
1077 | ||
1da177e4 LT |
1078 | /* |
1079 | parent == TC_H_UNSPEC - unspecified parent. | |
1080 | parent == TC_H_ROOT - class is root, which has no parent. | |
1081 | parent == X:0 - parent is root class. | |
1082 | parent == X:Y - parent is a node in hierarchy. | |
1083 | parent == 0:Y - parent is X:Y, where X:0 is qdisc. | |
1084 | ||
1085 | handle == 0:0 - generate handle from kernel pool. | |
1086 | handle == 0:Y - class is X:Y, where X:0 is qdisc. | |
1087 | handle == X:Y - clear. | |
1088 | handle == X:0 - root class. | |
1089 | */ | |
1090 | ||
1091 | /* Step 1. Determine qdisc handle X:0 */ | |
1092 | ||
1093 | if (pid != TC_H_ROOT) { | |
1094 | u32 qid1 = TC_H_MAJ(pid); | |
1095 | ||
1096 | if (qid && qid1) { | |
1097 | /* If both majors are known, they must be identical. */ | |
1098 | if (qid != qid1) | |
1099 | return -EINVAL; | |
1100 | } else if (qid1) { | |
1101 | qid = qid1; | |
1102 | } else if (qid == 0) | |
1103 | qid = dev->qdisc_sleeping->handle; | |
1104 | ||
1105 | /* Now qid is genuine qdisc handle consistent | |
1106 | both with parent and child. | |
1107 | ||
1108 | TC_H_MAJ(pid) still may be unspecified, complete it now. | |
1109 | */ | |
1110 | if (pid) | |
1111 | pid = TC_H_MAKE(qid, pid); | |
1112 | } else { | |
1113 | if (qid == 0) | |
1114 | qid = dev->qdisc_sleeping->handle; | |
1115 | } | |
1116 | ||
1117 | /* OK. Locate qdisc */ | |
10297b99 | 1118 | if ((q = qdisc_lookup(dev, qid)) == NULL) |
1da177e4 LT |
1119 | return -ENOENT; |
1120 | ||
1121 | /* An check that it supports classes */ | |
1122 | cops = q->ops->cl_ops; | |
1123 | if (cops == NULL) | |
1124 | return -EINVAL; | |
1125 | ||
1126 | /* Now try to get class */ | |
1127 | if (clid == 0) { | |
1128 | if (pid == TC_H_ROOT) | |
1129 | clid = qid; | |
1130 | } else | |
1131 | clid = TC_H_MAKE(qid, clid); | |
1132 | ||
1133 | if (clid) | |
1134 | cl = cops->get(q, clid); | |
1135 | ||
1136 | if (cl == 0) { | |
1137 | err = -ENOENT; | |
1138 | if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE)) | |
1139 | goto out; | |
1140 | } else { | |
1141 | switch (n->nlmsg_type) { | |
10297b99 | 1142 | case RTM_NEWTCLASS: |
1da177e4 LT |
1143 | err = -EEXIST; |
1144 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1145 | goto out; | |
1146 | break; | |
1147 | case RTM_DELTCLASS: | |
1148 | err = cops->delete(q, cl); | |
1149 | if (err == 0) | |
1150 | tclass_notify(skb, n, q, cl, RTM_DELTCLASS); | |
1151 | goto out; | |
1152 | case RTM_GETTCLASS: | |
1153 | err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS); | |
1154 | goto out; | |
1155 | default: | |
1156 | err = -EINVAL; | |
1157 | goto out; | |
1158 | } | |
1159 | } | |
1160 | ||
1161 | new_cl = cl; | |
1162 | err = cops->change(q, clid, pid, tca, &new_cl); | |
1163 | if (err == 0) | |
1164 | tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS); | |
1165 | ||
1166 | out: | |
1167 | if (cl) | |
1168 | cops->put(q, cl); | |
1169 | ||
1170 | return err; | |
1171 | } | |
1172 | ||
1173 | ||
1174 | static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, | |
1175 | unsigned long cl, | |
e431b8c0 | 1176 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
1177 | { |
1178 | struct tcmsg *tcm; | |
1179 | struct nlmsghdr *nlh; | |
27a884dc | 1180 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 | 1181 | struct gnet_dump d; |
20fea08b | 1182 | const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; |
1da177e4 | 1183 | |
e431b8c0 | 1184 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
1185 | tcm = NLMSG_DATA(nlh); |
1186 | tcm->tcm_family = AF_UNSPEC; | |
1187 | tcm->tcm_ifindex = q->dev->ifindex; | |
1188 | tcm->tcm_parent = q->handle; | |
1189 | tcm->tcm_handle = q->handle; | |
1190 | tcm->tcm_info = 0; | |
57e1c487 | 1191 | NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); |
1da177e4 | 1192 | if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) |
1e90474c | 1193 | goto nla_put_failure; |
1da177e4 LT |
1194 | |
1195 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
1196 | TCA_XSTATS, q->stats_lock, &d) < 0) | |
1e90474c | 1197 | goto nla_put_failure; |
1da177e4 LT |
1198 | |
1199 | if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) | |
1e90474c | 1200 | goto nla_put_failure; |
1da177e4 LT |
1201 | |
1202 | if (gnet_stats_finish_copy(&d) < 0) | |
1e90474c | 1203 | goto nla_put_failure; |
1da177e4 | 1204 | |
27a884dc | 1205 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
1206 | return skb->len; |
1207 | ||
1208 | nlmsg_failure: | |
1e90474c | 1209 | nla_put_failure: |
dc5fc579 | 1210 | nlmsg_trim(skb, b); |
1da177e4 LT |
1211 | return -1; |
1212 | } | |
1213 | ||
1214 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1215 | struct Qdisc *q, unsigned long cl, int event) | |
1216 | { | |
1217 | struct sk_buff *skb; | |
1218 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1219 | ||
1220 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1221 | if (!skb) | |
1222 | return -ENOBUFS; | |
1223 | ||
1224 | if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) { | |
1225 | kfree_skb(skb); | |
1226 | return -EINVAL; | |
1227 | } | |
1228 | ||
97c53cac | 1229 | return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
1230 | } |
1231 | ||
1232 | struct qdisc_dump_args | |
1233 | { | |
1234 | struct qdisc_walker w; | |
1235 | struct sk_buff *skb; | |
1236 | struct netlink_callback *cb; | |
1237 | }; | |
1238 | ||
1239 | static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) | |
1240 | { | |
1241 | struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; | |
1242 | ||
1243 | return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid, | |
1244 | a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); | |
1245 | } | |
1246 | ||
1247 | static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) | |
1248 | { | |
3b1e0a65 | 1249 | struct net *net = sock_net(skb->sk); |
1da177e4 LT |
1250 | int t; |
1251 | int s_t; | |
1252 | struct net_device *dev; | |
1253 | struct Qdisc *q; | |
1254 | struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); | |
1255 | struct qdisc_dump_args arg; | |
1256 | ||
b854272b DL |
1257 | if (net != &init_net) |
1258 | return 0; | |
1259 | ||
1da177e4 LT |
1260 | if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) |
1261 | return 0; | |
881d966b | 1262 | if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
1263 | return 0; |
1264 | ||
1265 | s_t = cb->args[0]; | |
1266 | t = 0; | |
1267 | ||
1da177e4 LT |
1268 | list_for_each_entry(q, &dev->qdisc_list, list) { |
1269 | if (t < s_t || !q->ops->cl_ops || | |
1270 | (tcm->tcm_parent && | |
1271 | TC_H_MAJ(tcm->tcm_parent) != q->handle)) { | |
1272 | t++; | |
1273 | continue; | |
1274 | } | |
1275 | if (t > s_t) | |
1276 | memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); | |
1277 | arg.w.fn = qdisc_class_dump; | |
1278 | arg.skb = skb; | |
1279 | arg.cb = cb; | |
1280 | arg.w.stop = 0; | |
1281 | arg.w.skip = cb->args[1]; | |
1282 | arg.w.count = 0; | |
1283 | q->ops->cl_ops->walk(q, &arg.w); | |
1284 | cb->args[1] = arg.w.count; | |
1285 | if (arg.w.stop) | |
1286 | break; | |
1287 | t++; | |
1288 | } | |
1da177e4 LT |
1289 | |
1290 | cb->args[0] = t; | |
1291 | ||
1292 | dev_put(dev); | |
1293 | return skb->len; | |
1294 | } | |
1295 | ||
1296 | /* Main classifier routine: scans classifier chain attached | |
1297 | to this qdisc, (optionally) tests for protocol and asks | |
1298 | specific classifiers. | |
1299 | */ | |
73ca4918 PM |
1300 | int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp, |
1301 | struct tcf_result *res) | |
1302 | { | |
1303 | __be16 protocol = skb->protocol; | |
1304 | int err = 0; | |
1305 | ||
1306 | for (; tp; tp = tp->next) { | |
1307 | if ((tp->protocol == protocol || | |
1308 | tp->protocol == htons(ETH_P_ALL)) && | |
1309 | (err = tp->classify(skb, tp, res)) >= 0) { | |
1310 | #ifdef CONFIG_NET_CLS_ACT | |
1311 | if (err != TC_ACT_RECLASSIFY && skb->tc_verd) | |
1312 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0); | |
1313 | #endif | |
1314 | return err; | |
1315 | } | |
1316 | } | |
1317 | return -1; | |
1318 | } | |
1319 | EXPORT_SYMBOL(tc_classify_compat); | |
1320 | ||
1da177e4 | 1321 | int tc_classify(struct sk_buff *skb, struct tcf_proto *tp, |
73ca4918 | 1322 | struct tcf_result *res) |
1da177e4 LT |
1323 | { |
1324 | int err = 0; | |
73ca4918 | 1325 | __be16 protocol; |
1da177e4 LT |
1326 | #ifdef CONFIG_NET_CLS_ACT |
1327 | struct tcf_proto *otp = tp; | |
1328 | reclassify: | |
1329 | #endif | |
1330 | protocol = skb->protocol; | |
1331 | ||
73ca4918 | 1332 | err = tc_classify_compat(skb, tp, res); |
1da177e4 | 1333 | #ifdef CONFIG_NET_CLS_ACT |
73ca4918 PM |
1334 | if (err == TC_ACT_RECLASSIFY) { |
1335 | u32 verd = G_TC_VERD(skb->tc_verd); | |
1336 | tp = otp; | |
1337 | ||
1338 | if (verd++ >= MAX_REC_LOOP) { | |
1339 | printk("rule prio %u protocol %02x reclassify loop, " | |
1340 | "packet dropped\n", | |
1341 | tp->prio&0xffff, ntohs(tp->protocol)); | |
1342 | return TC_ACT_SHOT; | |
1da177e4 | 1343 | } |
73ca4918 PM |
1344 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd); |
1345 | goto reclassify; | |
1da177e4 | 1346 | } |
73ca4918 PM |
1347 | #endif |
1348 | return err; | |
1da177e4 | 1349 | } |
73ca4918 | 1350 | EXPORT_SYMBOL(tc_classify); |
1da177e4 | 1351 | |
a48b5a61 PM |
1352 | void tcf_destroy(struct tcf_proto *tp) |
1353 | { | |
1354 | tp->ops->destroy(tp); | |
1355 | module_put(tp->ops->owner); | |
1356 | kfree(tp); | |
1357 | } | |
1358 | ||
ff31ab56 | 1359 | void tcf_destroy_chain(struct tcf_proto **fl) |
a48b5a61 PM |
1360 | { |
1361 | struct tcf_proto *tp; | |
1362 | ||
ff31ab56 PM |
1363 | while ((tp = *fl) != NULL) { |
1364 | *fl = tp->next; | |
a48b5a61 PM |
1365 | tcf_destroy(tp); |
1366 | } | |
1367 | } | |
1368 | EXPORT_SYMBOL(tcf_destroy_chain); | |
1369 | ||
1da177e4 LT |
1370 | #ifdef CONFIG_PROC_FS |
1371 | static int psched_show(struct seq_file *seq, void *v) | |
1372 | { | |
3c0cfc13 PM |
1373 | struct timespec ts; |
1374 | ||
1375 | hrtimer_get_res(CLOCK_MONOTONIC, &ts); | |
1da177e4 | 1376 | seq_printf(seq, "%08x %08x %08x %08x\n", |
641b9e0e | 1377 | (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1), |
514bca32 | 1378 | 1000000, |
3c0cfc13 | 1379 | (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts))); |
1da177e4 LT |
1380 | |
1381 | return 0; | |
1382 | } | |
1383 | ||
1384 | static int psched_open(struct inode *inode, struct file *file) | |
1385 | { | |
1386 | return single_open(file, psched_show, PDE(inode)->data); | |
1387 | } | |
1388 | ||
da7071d7 | 1389 | static const struct file_operations psched_fops = { |
1da177e4 LT |
1390 | .owner = THIS_MODULE, |
1391 | .open = psched_open, | |
1392 | .read = seq_read, | |
1393 | .llseek = seq_lseek, | |
1394 | .release = single_release, | |
10297b99 | 1395 | }; |
1da177e4 LT |
1396 | #endif |
1397 | ||
1da177e4 LT |
1398 | static int __init pktsched_init(void) |
1399 | { | |
1da177e4 LT |
1400 | register_qdisc(&pfifo_qdisc_ops); |
1401 | register_qdisc(&bfifo_qdisc_ops); | |
457c4cbc | 1402 | proc_net_fops_create(&init_net, "psched", 0, &psched_fops); |
1da177e4 | 1403 | |
be577ddc TG |
1404 | rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL); |
1405 | rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL); | |
1406 | rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc); | |
1407 | rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL); | |
1408 | rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL); | |
1409 | rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass); | |
1410 | ||
1da177e4 LT |
1411 | return 0; |
1412 | } | |
1413 | ||
1414 | subsys_initcall(pktsched_init); |