]>
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 | |
e65d22e1 | 102 | just buggy devices, which can defer output even if netif_queue_stopped()=0. |
1da177e4 LT |
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 | 187 | { |
b0e1e646 | 188 | struct netdev_queue *dev_queue = &dev->tx_queue; |
1da177e4 LT |
189 | struct Qdisc *q; |
190 | ||
b0e1e646 | 191 | list_for_each_entry(q, &dev_queue->qdisc_list, list) { |
43effa1e | 192 | if (q->handle == handle) |
1da177e4 | 193 | return q; |
1da177e4 | 194 | } |
1da177e4 LT |
195 | return NULL; |
196 | } | |
197 | ||
198 | static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) | |
199 | { | |
200 | unsigned long cl; | |
201 | struct Qdisc *leaf; | |
20fea08b | 202 | const struct Qdisc_class_ops *cops = p->ops->cl_ops; |
1da177e4 LT |
203 | |
204 | if (cops == NULL) | |
205 | return NULL; | |
206 | cl = cops->get(p, classid); | |
207 | ||
208 | if (cl == 0) | |
209 | return NULL; | |
210 | leaf = cops->leaf(p, cl); | |
211 | cops->put(p, cl); | |
212 | return leaf; | |
213 | } | |
214 | ||
215 | /* Find queueing discipline by name */ | |
216 | ||
1e90474c | 217 | static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) |
1da177e4 LT |
218 | { |
219 | struct Qdisc_ops *q = NULL; | |
220 | ||
221 | if (kind) { | |
222 | read_lock(&qdisc_mod_lock); | |
223 | for (q = qdisc_base; q; q = q->next) { | |
1e90474c | 224 | if (nla_strcmp(kind, q->id) == 0) { |
1da177e4 LT |
225 | if (!try_module_get(q->owner)) |
226 | q = NULL; | |
227 | break; | |
228 | } | |
229 | } | |
230 | read_unlock(&qdisc_mod_lock); | |
231 | } | |
232 | return q; | |
233 | } | |
234 | ||
235 | static struct qdisc_rate_table *qdisc_rtab_list; | |
236 | ||
1e90474c | 237 | struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab) |
1da177e4 LT |
238 | { |
239 | struct qdisc_rate_table *rtab; | |
240 | ||
241 | for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { | |
242 | if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) { | |
243 | rtab->refcnt++; | |
244 | return rtab; | |
245 | } | |
246 | } | |
247 | ||
5feb5e1a PM |
248 | if (tab == NULL || r->rate == 0 || r->cell_log == 0 || |
249 | nla_len(tab) != TC_RTAB_SIZE) | |
1da177e4 LT |
250 | return NULL; |
251 | ||
252 | rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); | |
253 | if (rtab) { | |
254 | rtab->rate = *r; | |
255 | rtab->refcnt = 1; | |
1e90474c | 256 | memcpy(rtab->data, nla_data(tab), 1024); |
1da177e4 LT |
257 | rtab->next = qdisc_rtab_list; |
258 | qdisc_rtab_list = rtab; | |
259 | } | |
260 | return rtab; | |
261 | } | |
62e3ba1b | 262 | EXPORT_SYMBOL(qdisc_get_rtab); |
1da177e4 LT |
263 | |
264 | void qdisc_put_rtab(struct qdisc_rate_table *tab) | |
265 | { | |
266 | struct qdisc_rate_table *rtab, **rtabp; | |
267 | ||
268 | if (!tab || --tab->refcnt) | |
269 | return; | |
270 | ||
271 | for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) { | |
272 | if (rtab == tab) { | |
273 | *rtabp = rtab->next; | |
274 | kfree(rtab); | |
275 | return; | |
276 | } | |
277 | } | |
278 | } | |
62e3ba1b | 279 | EXPORT_SYMBOL(qdisc_put_rtab); |
1da177e4 | 280 | |
4179477f PM |
281 | static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) |
282 | { | |
283 | struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, | |
284 | timer); | |
5ce2d488 | 285 | struct net_device *dev = qdisc_dev(wd->qdisc); |
4179477f PM |
286 | |
287 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
11274e5a | 288 | smp_wmb(); |
0621ed2e | 289 | netif_schedule(dev); |
1936502d | 290 | |
4179477f PM |
291 | return HRTIMER_NORESTART; |
292 | } | |
293 | ||
294 | void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) | |
295 | { | |
296 | hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
297 | wd->timer.function = qdisc_watchdog; | |
298 | wd->qdisc = qdisc; | |
299 | } | |
300 | EXPORT_SYMBOL(qdisc_watchdog_init); | |
301 | ||
302 | void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires) | |
303 | { | |
304 | ktime_t time; | |
305 | ||
306 | wd->qdisc->flags |= TCQ_F_THROTTLED; | |
307 | time = ktime_set(0, 0); | |
308 | time = ktime_add_ns(time, PSCHED_US2NS(expires)); | |
309 | hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS); | |
310 | } | |
311 | EXPORT_SYMBOL(qdisc_watchdog_schedule); | |
312 | ||
313 | void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) | |
314 | { | |
315 | hrtimer_cancel(&wd->timer); | |
316 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
317 | } | |
318 | EXPORT_SYMBOL(qdisc_watchdog_cancel); | |
1da177e4 | 319 | |
6fe1c7a5 PM |
320 | struct hlist_head *qdisc_class_hash_alloc(unsigned int n) |
321 | { | |
322 | unsigned int size = n * sizeof(struct hlist_head), i; | |
323 | struct hlist_head *h; | |
324 | ||
325 | if (size <= PAGE_SIZE) | |
326 | h = kmalloc(size, GFP_KERNEL); | |
327 | else | |
328 | h = (struct hlist_head *) | |
329 | __get_free_pages(GFP_KERNEL, get_order(size)); | |
330 | ||
331 | if (h != NULL) { | |
332 | for (i = 0; i < n; i++) | |
333 | INIT_HLIST_HEAD(&h[i]); | |
334 | } | |
335 | return h; | |
336 | } | |
337 | ||
338 | static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) | |
339 | { | |
340 | unsigned int size = n * sizeof(struct hlist_head); | |
341 | ||
342 | if (size <= PAGE_SIZE) | |
343 | kfree(h); | |
344 | else | |
345 | free_pages((unsigned long)h, get_order(size)); | |
346 | } | |
347 | ||
348 | void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) | |
349 | { | |
350 | struct Qdisc_class_common *cl; | |
351 | struct hlist_node *n, *next; | |
352 | struct hlist_head *nhash, *ohash; | |
353 | unsigned int nsize, nmask, osize; | |
354 | unsigned int i, h; | |
355 | ||
356 | /* Rehash when load factor exceeds 0.75 */ | |
357 | if (clhash->hashelems * 4 <= clhash->hashsize * 3) | |
358 | return; | |
359 | nsize = clhash->hashsize * 2; | |
360 | nmask = nsize - 1; | |
361 | nhash = qdisc_class_hash_alloc(nsize); | |
362 | if (nhash == NULL) | |
363 | return; | |
364 | ||
365 | ohash = clhash->hash; | |
366 | osize = clhash->hashsize; | |
367 | ||
368 | sch_tree_lock(sch); | |
369 | for (i = 0; i < osize; i++) { | |
370 | hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) { | |
371 | h = qdisc_class_hash(cl->classid, nmask); | |
372 | hlist_add_head(&cl->hnode, &nhash[h]); | |
373 | } | |
374 | } | |
375 | clhash->hash = nhash; | |
376 | clhash->hashsize = nsize; | |
377 | clhash->hashmask = nmask; | |
378 | sch_tree_unlock(sch); | |
379 | ||
380 | qdisc_class_hash_free(ohash, osize); | |
381 | } | |
382 | EXPORT_SYMBOL(qdisc_class_hash_grow); | |
383 | ||
384 | int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) | |
385 | { | |
386 | unsigned int size = 4; | |
387 | ||
388 | clhash->hash = qdisc_class_hash_alloc(size); | |
389 | if (clhash->hash == NULL) | |
390 | return -ENOMEM; | |
391 | clhash->hashsize = size; | |
392 | clhash->hashmask = size - 1; | |
393 | clhash->hashelems = 0; | |
394 | return 0; | |
395 | } | |
396 | EXPORT_SYMBOL(qdisc_class_hash_init); | |
397 | ||
398 | void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) | |
399 | { | |
400 | qdisc_class_hash_free(clhash->hash, clhash->hashsize); | |
401 | } | |
402 | EXPORT_SYMBOL(qdisc_class_hash_destroy); | |
403 | ||
404 | void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, | |
405 | struct Qdisc_class_common *cl) | |
406 | { | |
407 | unsigned int h; | |
408 | ||
409 | INIT_HLIST_NODE(&cl->hnode); | |
410 | h = qdisc_class_hash(cl->classid, clhash->hashmask); | |
411 | hlist_add_head(&cl->hnode, &clhash->hash[h]); | |
412 | clhash->hashelems++; | |
413 | } | |
414 | EXPORT_SYMBOL(qdisc_class_hash_insert); | |
415 | ||
416 | void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, | |
417 | struct Qdisc_class_common *cl) | |
418 | { | |
419 | hlist_del(&cl->hnode); | |
420 | clhash->hashelems--; | |
421 | } | |
422 | EXPORT_SYMBOL(qdisc_class_hash_remove); | |
423 | ||
1da177e4 LT |
424 | /* Allocate an unique handle from space managed by kernel */ |
425 | ||
426 | static u32 qdisc_alloc_handle(struct net_device *dev) | |
427 | { | |
428 | int i = 0x10000; | |
429 | static u32 autohandle = TC_H_MAKE(0x80000000U, 0); | |
430 | ||
431 | do { | |
432 | autohandle += TC_H_MAKE(0x10000U, 0); | |
433 | if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) | |
434 | autohandle = TC_H_MAKE(0x80000000U, 0); | |
435 | } while (qdisc_lookup(dev, autohandle) && --i > 0); | |
436 | ||
437 | return i>0 ? autohandle : 0; | |
438 | } | |
439 | ||
440 | /* Attach toplevel qdisc to device dev */ | |
441 | ||
442 | static struct Qdisc * | |
443 | dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc) | |
444 | { | |
b0e1e646 | 445 | struct netdev_queue *dev_queue; |
1da177e4 LT |
446 | struct Qdisc *oqdisc; |
447 | ||
448 | if (dev->flags & IFF_UP) | |
449 | dev_deactivate(dev); | |
450 | ||
451 | qdisc_lock_tree(dev); | |
452 | if (qdisc && qdisc->flags&TCQ_F_INGRESS) { | |
453 | oqdisc = dev->qdisc_ingress; | |
454 | /* Prune old scheduler */ | |
455 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) { | |
456 | /* delete */ | |
457 | qdisc_reset(oqdisc); | |
458 | dev->qdisc_ingress = NULL; | |
459 | } else { /* new */ | |
460 | dev->qdisc_ingress = qdisc; | |
461 | } | |
462 | ||
463 | } else { | |
b0e1e646 DM |
464 | dev_queue = &dev->tx_queue; |
465 | oqdisc = dev_queue->qdisc_sleeping; | |
1da177e4 LT |
466 | |
467 | /* Prune old scheduler */ | |
468 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
469 | qdisc_reset(oqdisc); | |
470 | ||
471 | /* ... and graft new one */ | |
472 | if (qdisc == NULL) | |
473 | qdisc = &noop_qdisc; | |
b0e1e646 DM |
474 | dev_queue->qdisc_sleeping = qdisc; |
475 | dev_queue->qdisc = &noop_qdisc; | |
1da177e4 LT |
476 | } |
477 | ||
478 | qdisc_unlock_tree(dev); | |
479 | ||
480 | if (dev->flags & IFF_UP) | |
481 | dev_activate(dev); | |
482 | ||
483 | return oqdisc; | |
484 | } | |
485 | ||
43effa1e PM |
486 | void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n) |
487 | { | |
20fea08b | 488 | const struct Qdisc_class_ops *cops; |
43effa1e PM |
489 | unsigned long cl; |
490 | u32 parentid; | |
491 | ||
492 | if (n == 0) | |
493 | return; | |
494 | while ((parentid = sch->parent)) { | |
066a3b5b JP |
495 | if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) |
496 | return; | |
497 | ||
5ce2d488 | 498 | sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); |
ffc8fefa PM |
499 | if (sch == NULL) { |
500 | WARN_ON(parentid != TC_H_ROOT); | |
501 | return; | |
502 | } | |
43effa1e PM |
503 | cops = sch->ops->cl_ops; |
504 | if (cops->qlen_notify) { | |
505 | cl = cops->get(sch, parentid); | |
506 | cops->qlen_notify(sch, cl); | |
507 | cops->put(sch, cl); | |
508 | } | |
509 | sch->q.qlen -= n; | |
510 | } | |
511 | } | |
512 | EXPORT_SYMBOL(qdisc_tree_decrease_qlen); | |
1da177e4 LT |
513 | |
514 | /* Graft qdisc "new" to class "classid" of qdisc "parent" or | |
515 | to device "dev". | |
516 | ||
517 | Old qdisc is not destroyed but returned in *old. | |
518 | */ | |
519 | ||
520 | static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, | |
521 | u32 classid, | |
522 | struct Qdisc *new, struct Qdisc **old) | |
523 | { | |
524 | int err = 0; | |
525 | struct Qdisc *q = *old; | |
526 | ||
527 | ||
10297b99 | 528 | if (parent == NULL) { |
1da177e4 LT |
529 | if (q && q->flags&TCQ_F_INGRESS) { |
530 | *old = dev_graft_qdisc(dev, q); | |
531 | } else { | |
532 | *old = dev_graft_qdisc(dev, new); | |
533 | } | |
534 | } else { | |
20fea08b | 535 | const struct Qdisc_class_ops *cops = parent->ops->cl_ops; |
1da177e4 LT |
536 | |
537 | err = -EINVAL; | |
538 | ||
539 | if (cops) { | |
540 | unsigned long cl = cops->get(parent, classid); | |
541 | if (cl) { | |
542 | err = cops->graft(parent, cl, new, old); | |
1da177e4 LT |
543 | cops->put(parent, cl); |
544 | } | |
545 | } | |
546 | } | |
547 | return err; | |
548 | } | |
549 | ||
550 | /* | |
551 | Allocate and initialize new qdisc. | |
552 | ||
553 | Parameters are passed via opt. | |
554 | */ | |
555 | ||
556 | static struct Qdisc * | |
bb949fbd DM |
557 | qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue, |
558 | u32 parent, u32 handle, struct nlattr **tca, int *errp) | |
1da177e4 LT |
559 | { |
560 | int err; | |
1e90474c | 561 | struct nlattr *kind = tca[TCA_KIND]; |
1da177e4 LT |
562 | struct Qdisc *sch; |
563 | struct Qdisc_ops *ops; | |
1da177e4 LT |
564 | |
565 | ops = qdisc_lookup_ops(kind); | |
566 | #ifdef CONFIG_KMOD | |
567 | if (ops == NULL && kind != NULL) { | |
568 | char name[IFNAMSIZ]; | |
1e90474c | 569 | if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { |
1da177e4 LT |
570 | /* We dropped the RTNL semaphore in order to |
571 | * perform the module load. So, even if we | |
572 | * succeeded in loading the module we have to | |
573 | * tell the caller to replay the request. We | |
574 | * indicate this using -EAGAIN. | |
575 | * We replay the request because the device may | |
576 | * go away in the mean time. | |
577 | */ | |
578 | rtnl_unlock(); | |
579 | request_module("sch_%s", name); | |
580 | rtnl_lock(); | |
581 | ops = qdisc_lookup_ops(kind); | |
582 | if (ops != NULL) { | |
583 | /* We will try again qdisc_lookup_ops, | |
584 | * so don't keep a reference. | |
585 | */ | |
586 | module_put(ops->owner); | |
587 | err = -EAGAIN; | |
588 | goto err_out; | |
589 | } | |
590 | } | |
591 | } | |
592 | #endif | |
593 | ||
b9e2cc0f | 594 | err = -ENOENT; |
1da177e4 LT |
595 | if (ops == NULL) |
596 | goto err_out; | |
597 | ||
5ce2d488 | 598 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
599 | if (IS_ERR(sch)) { |
600 | err = PTR_ERR(sch); | |
1da177e4 | 601 | goto err_out2; |
3d54b82f | 602 | } |
1da177e4 | 603 | |
ffc8fefa PM |
604 | sch->parent = parent; |
605 | ||
555353cf | 606 | sch->stats_lock = &dev_queue->lock; |
3d54b82f | 607 | if (handle == TC_H_INGRESS) { |
1da177e4 | 608 | sch->flags |= TCQ_F_INGRESS; |
3d54b82f | 609 | handle = TC_H_MAKE(TC_H_INGRESS, 0); |
fd44de7c | 610 | } else { |
fd44de7c PM |
611 | if (handle == 0) { |
612 | handle = qdisc_alloc_handle(dev); | |
613 | err = -ENOMEM; | |
614 | if (handle == 0) | |
615 | goto err_out3; | |
616 | } | |
1da177e4 LT |
617 | } |
618 | ||
3d54b82f | 619 | sch->handle = handle; |
1da177e4 | 620 | |
1e90474c PM |
621 | if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { |
622 | if (tca[TCA_RATE]) { | |
023e09a7 TG |
623 | err = gen_new_estimator(&sch->bstats, &sch->rate_est, |
624 | sch->stats_lock, | |
1e90474c | 625 | tca[TCA_RATE]); |
023e09a7 TG |
626 | if (err) { |
627 | /* | |
628 | * Any broken qdiscs that would require | |
629 | * a ops->reset() here? The qdisc was never | |
630 | * in action so it shouldn't be necessary. | |
631 | */ | |
632 | if (ops->destroy) | |
633 | ops->destroy(sch); | |
634 | goto err_out3; | |
635 | } | |
636 | } | |
1da177e4 | 637 | qdisc_lock_tree(dev); |
b0e1e646 | 638 | list_add_tail(&sch->list, &dev_queue->qdisc_list); |
1da177e4 LT |
639 | qdisc_unlock_tree(dev); |
640 | ||
1da177e4 LT |
641 | return sch; |
642 | } | |
643 | err_out3: | |
644 | dev_put(dev); | |
3d54b82f | 645 | kfree((char *) sch - sch->padded); |
1da177e4 LT |
646 | err_out2: |
647 | module_put(ops->owner); | |
648 | err_out: | |
649 | *errp = err; | |
1da177e4 LT |
650 | return NULL; |
651 | } | |
652 | ||
1e90474c | 653 | static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) |
1da177e4 | 654 | { |
1e90474c | 655 | if (tca[TCA_OPTIONS]) { |
1da177e4 LT |
656 | int err; |
657 | ||
658 | if (sch->ops->change == NULL) | |
659 | return -EINVAL; | |
1e90474c | 660 | err = sch->ops->change(sch, tca[TCA_OPTIONS]); |
1da177e4 LT |
661 | if (err) |
662 | return err; | |
663 | } | |
1e90474c | 664 | if (tca[TCA_RATE]) |
1da177e4 | 665 | gen_replace_estimator(&sch->bstats, &sch->rate_est, |
1e90474c | 666 | sch->stats_lock, tca[TCA_RATE]); |
1da177e4 LT |
667 | return 0; |
668 | } | |
669 | ||
670 | struct check_loop_arg | |
671 | { | |
672 | struct qdisc_walker w; | |
673 | struct Qdisc *p; | |
674 | int depth; | |
675 | }; | |
676 | ||
677 | static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); | |
678 | ||
679 | static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) | |
680 | { | |
681 | struct check_loop_arg arg; | |
682 | ||
683 | if (q->ops->cl_ops == NULL) | |
684 | return 0; | |
685 | ||
686 | arg.w.stop = arg.w.skip = arg.w.count = 0; | |
687 | arg.w.fn = check_loop_fn; | |
688 | arg.depth = depth; | |
689 | arg.p = p; | |
690 | q->ops->cl_ops->walk(q, &arg.w); | |
691 | return arg.w.stop ? -ELOOP : 0; | |
692 | } | |
693 | ||
694 | static int | |
695 | check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) | |
696 | { | |
697 | struct Qdisc *leaf; | |
20fea08b | 698 | const struct Qdisc_class_ops *cops = q->ops->cl_ops; |
1da177e4 LT |
699 | struct check_loop_arg *arg = (struct check_loop_arg *)w; |
700 | ||
701 | leaf = cops->leaf(q, cl); | |
702 | if (leaf) { | |
703 | if (leaf == arg->p || arg->depth > 7) | |
704 | return -ELOOP; | |
705 | return check_loop(leaf, arg->p, arg->depth + 1); | |
706 | } | |
707 | return 0; | |
708 | } | |
709 | ||
710 | /* | |
711 | * Delete/get qdisc. | |
712 | */ | |
713 | ||
714 | static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
715 | { | |
3b1e0a65 | 716 | struct net *net = sock_net(skb->sk); |
1da177e4 | 717 | struct tcmsg *tcm = NLMSG_DATA(n); |
1e90474c | 718 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
719 | struct net_device *dev; |
720 | u32 clid = tcm->tcm_parent; | |
721 | struct Qdisc *q = NULL; | |
722 | struct Qdisc *p = NULL; | |
723 | int err; | |
724 | ||
b854272b DL |
725 | if (net != &init_net) |
726 | return -EINVAL; | |
727 | ||
881d966b | 728 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
729 | return -ENODEV; |
730 | ||
1e90474c PM |
731 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); |
732 | if (err < 0) | |
733 | return err; | |
734 | ||
1da177e4 LT |
735 | if (clid) { |
736 | if (clid != TC_H_ROOT) { | |
737 | if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { | |
738 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
739 | return -ENOENT; | |
740 | q = qdisc_leaf(p, clid); | |
741 | } else { /* ingress */ | |
742 | q = dev->qdisc_ingress; | |
10297b99 | 743 | } |
1da177e4 | 744 | } else { |
b0e1e646 DM |
745 | struct netdev_queue *dev_queue = &dev->tx_queue; |
746 | q = dev_queue->qdisc_sleeping; | |
1da177e4 LT |
747 | } |
748 | if (!q) | |
749 | return -ENOENT; | |
750 | ||
751 | if (tcm->tcm_handle && q->handle != tcm->tcm_handle) | |
752 | return -EINVAL; | |
753 | } else { | |
754 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
755 | return -ENOENT; | |
756 | } | |
757 | ||
1e90474c | 758 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
759 | return -EINVAL; |
760 | ||
761 | if (n->nlmsg_type == RTM_DELQDISC) { | |
762 | if (!clid) | |
763 | return -EINVAL; | |
764 | if (q->handle == 0) | |
765 | return -ENOENT; | |
766 | if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0) | |
767 | return err; | |
768 | if (q) { | |
769 | qdisc_notify(skb, n, clid, q, NULL); | |
fd44de7c | 770 | qdisc_lock_tree(dev); |
1da177e4 | 771 | qdisc_destroy(q); |
fd44de7c | 772 | qdisc_unlock_tree(dev); |
1da177e4 LT |
773 | } |
774 | } else { | |
775 | qdisc_notify(skb, n, clid, NULL, q); | |
776 | } | |
777 | return 0; | |
778 | } | |
779 | ||
780 | /* | |
781 | Create/change qdisc. | |
782 | */ | |
783 | ||
784 | static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
785 | { | |
3b1e0a65 | 786 | struct net *net = sock_net(skb->sk); |
1da177e4 | 787 | struct tcmsg *tcm; |
1e90474c | 788 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
789 | struct net_device *dev; |
790 | u32 clid; | |
791 | struct Qdisc *q, *p; | |
792 | int err; | |
793 | ||
b854272b DL |
794 | if (net != &init_net) |
795 | return -EINVAL; | |
796 | ||
1da177e4 LT |
797 | replay: |
798 | /* Reinit, just in case something touches this. */ | |
799 | tcm = NLMSG_DATA(n); | |
1da177e4 LT |
800 | clid = tcm->tcm_parent; |
801 | q = p = NULL; | |
802 | ||
881d966b | 803 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
804 | return -ENODEV; |
805 | ||
1e90474c PM |
806 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); |
807 | if (err < 0) | |
808 | return err; | |
809 | ||
1da177e4 LT |
810 | if (clid) { |
811 | if (clid != TC_H_ROOT) { | |
812 | if (clid != TC_H_INGRESS) { | |
813 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
814 | return -ENOENT; | |
815 | q = qdisc_leaf(p, clid); | |
816 | } else { /*ingress */ | |
817 | q = dev->qdisc_ingress; | |
818 | } | |
819 | } else { | |
b0e1e646 DM |
820 | struct netdev_queue *dev_queue = &dev->tx_queue; |
821 | q = dev_queue->qdisc_sleeping; | |
1da177e4 LT |
822 | } |
823 | ||
824 | /* It may be default qdisc, ignore it */ | |
825 | if (q && q->handle == 0) | |
826 | q = NULL; | |
827 | ||
828 | if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { | |
829 | if (tcm->tcm_handle) { | |
830 | if (q && !(n->nlmsg_flags&NLM_F_REPLACE)) | |
831 | return -EEXIST; | |
832 | if (TC_H_MIN(tcm->tcm_handle)) | |
833 | return -EINVAL; | |
834 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
835 | goto create_n_graft; | |
836 | if (n->nlmsg_flags&NLM_F_EXCL) | |
837 | return -EEXIST; | |
1e90474c | 838 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
839 | return -EINVAL; |
840 | if (q == p || | |
841 | (p && check_loop(q, p, 0))) | |
842 | return -ELOOP; | |
843 | atomic_inc(&q->refcnt); | |
844 | goto graft; | |
845 | } else { | |
846 | if (q == NULL) | |
847 | goto create_n_graft; | |
848 | ||
849 | /* This magic test requires explanation. | |
850 | * | |
851 | * We know, that some child q is already | |
852 | * attached to this parent and have choice: | |
853 | * either to change it or to create/graft new one. | |
854 | * | |
855 | * 1. We are allowed to create/graft only | |
856 | * if CREATE and REPLACE flags are set. | |
857 | * | |
858 | * 2. If EXCL is set, requestor wanted to say, | |
859 | * that qdisc tcm_handle is not expected | |
860 | * to exist, so that we choose create/graft too. | |
861 | * | |
862 | * 3. The last case is when no flags are set. | |
863 | * Alas, it is sort of hole in API, we | |
864 | * cannot decide what to do unambiguously. | |
865 | * For now we select create/graft, if | |
866 | * user gave KIND, which does not match existing. | |
867 | */ | |
868 | if ((n->nlmsg_flags&NLM_F_CREATE) && | |
869 | (n->nlmsg_flags&NLM_F_REPLACE) && | |
870 | ((n->nlmsg_flags&NLM_F_EXCL) || | |
1e90474c PM |
871 | (tca[TCA_KIND] && |
872 | nla_strcmp(tca[TCA_KIND], q->ops->id)))) | |
1da177e4 LT |
873 | goto create_n_graft; |
874 | } | |
875 | } | |
876 | } else { | |
877 | if (!tcm->tcm_handle) | |
878 | return -EINVAL; | |
879 | q = qdisc_lookup(dev, tcm->tcm_handle); | |
880 | } | |
881 | ||
882 | /* Change qdisc parameters */ | |
883 | if (q == NULL) | |
884 | return -ENOENT; | |
885 | if (n->nlmsg_flags&NLM_F_EXCL) | |
886 | return -EEXIST; | |
1e90474c | 887 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
888 | return -EINVAL; |
889 | err = qdisc_change(q, tca); | |
890 | if (err == 0) | |
891 | qdisc_notify(skb, n, clid, NULL, q); | |
892 | return err; | |
893 | ||
894 | create_n_graft: | |
895 | if (!(n->nlmsg_flags&NLM_F_CREATE)) | |
896 | return -ENOENT; | |
897 | if (clid == TC_H_INGRESS) | |
bb949fbd DM |
898 | q = qdisc_create(dev, &dev->rx_queue, |
899 | tcm->tcm_parent, tcm->tcm_parent, | |
ffc8fefa | 900 | tca, &err); |
10297b99 | 901 | else |
bb949fbd DM |
902 | q = qdisc_create(dev, &dev->tx_queue, |
903 | tcm->tcm_parent, tcm->tcm_handle, | |
ffc8fefa | 904 | tca, &err); |
1da177e4 LT |
905 | if (q == NULL) { |
906 | if (err == -EAGAIN) | |
907 | goto replay; | |
908 | return err; | |
909 | } | |
910 | ||
911 | graft: | |
912 | if (1) { | |
913 | struct Qdisc *old_q = NULL; | |
914 | err = qdisc_graft(dev, p, clid, q, &old_q); | |
915 | if (err) { | |
916 | if (q) { | |
fd44de7c | 917 | qdisc_lock_tree(dev); |
1da177e4 | 918 | qdisc_destroy(q); |
fd44de7c | 919 | qdisc_unlock_tree(dev); |
1da177e4 LT |
920 | } |
921 | return err; | |
922 | } | |
923 | qdisc_notify(skb, n, clid, old_q, q); | |
924 | if (old_q) { | |
fd44de7c | 925 | qdisc_lock_tree(dev); |
1da177e4 | 926 | qdisc_destroy(old_q); |
fd44de7c | 927 | qdisc_unlock_tree(dev); |
1da177e4 LT |
928 | } |
929 | } | |
930 | return 0; | |
931 | } | |
932 | ||
933 | static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, | |
e431b8c0 | 934 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
935 | { |
936 | struct tcmsg *tcm; | |
937 | struct nlmsghdr *nlh; | |
27a884dc | 938 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
939 | struct gnet_dump d; |
940 | ||
e431b8c0 | 941 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
942 | tcm = NLMSG_DATA(nlh); |
943 | tcm->tcm_family = AF_UNSPEC; | |
9ef1d4c7 PM |
944 | tcm->tcm__pad1 = 0; |
945 | tcm->tcm__pad2 = 0; | |
5ce2d488 | 946 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex; |
1da177e4 LT |
947 | tcm->tcm_parent = clid; |
948 | tcm->tcm_handle = q->handle; | |
949 | tcm->tcm_info = atomic_read(&q->refcnt); | |
57e1c487 | 950 | NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); |
1da177e4 | 951 | if (q->ops->dump && q->ops->dump(q, skb) < 0) |
1e90474c | 952 | goto nla_put_failure; |
1da177e4 LT |
953 | q->qstats.qlen = q->q.qlen; |
954 | ||
955 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
956 | TCA_XSTATS, q->stats_lock, &d) < 0) | |
1e90474c | 957 | goto nla_put_failure; |
1da177e4 LT |
958 | |
959 | if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) | |
1e90474c | 960 | goto nla_put_failure; |
1da177e4 LT |
961 | |
962 | if (gnet_stats_copy_basic(&d, &q->bstats) < 0 || | |
1da177e4 | 963 | gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || |
1da177e4 | 964 | gnet_stats_copy_queue(&d, &q->qstats) < 0) |
1e90474c | 965 | goto nla_put_failure; |
10297b99 | 966 | |
1da177e4 | 967 | if (gnet_stats_finish_copy(&d) < 0) |
1e90474c | 968 | goto nla_put_failure; |
10297b99 | 969 | |
27a884dc | 970 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
971 | return skb->len; |
972 | ||
973 | nlmsg_failure: | |
1e90474c | 974 | nla_put_failure: |
dc5fc579 | 975 | nlmsg_trim(skb, b); |
1da177e4 LT |
976 | return -1; |
977 | } | |
978 | ||
979 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
980 | u32 clid, struct Qdisc *old, struct Qdisc *new) | |
981 | { | |
982 | struct sk_buff *skb; | |
983 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
984 | ||
985 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
986 | if (!skb) | |
987 | return -ENOBUFS; | |
988 | ||
989 | if (old && old->handle) { | |
990 | if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0) | |
991 | goto err_out; | |
992 | } | |
993 | if (new) { | |
994 | if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) | |
995 | goto err_out; | |
996 | } | |
997 | ||
998 | if (skb->len) | |
97c53cac | 999 | return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
1000 | |
1001 | err_out: | |
1002 | kfree_skb(skb); | |
1003 | return -EINVAL; | |
1004 | } | |
1005 | ||
1006 | static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) | |
1007 | { | |
3b1e0a65 | 1008 | struct net *net = sock_net(skb->sk); |
1da177e4 LT |
1009 | int idx, q_idx; |
1010 | int s_idx, s_q_idx; | |
1011 | struct net_device *dev; | |
1012 | struct Qdisc *q; | |
1013 | ||
b854272b DL |
1014 | if (net != &init_net) |
1015 | return 0; | |
1016 | ||
1da177e4 LT |
1017 | s_idx = cb->args[0]; |
1018 | s_q_idx = q_idx = cb->args[1]; | |
1019 | read_lock(&dev_base_lock); | |
7562f876 | 1020 | idx = 0; |
881d966b | 1021 | for_each_netdev(&init_net, dev) { |
b0e1e646 | 1022 | struct netdev_queue *dev_queue; |
1da177e4 | 1023 | if (idx < s_idx) |
7562f876 | 1024 | goto cont; |
1da177e4 LT |
1025 | if (idx > s_idx) |
1026 | s_q_idx = 0; | |
1da177e4 | 1027 | q_idx = 0; |
b0e1e646 DM |
1028 | dev_queue = &dev->tx_queue; |
1029 | list_for_each_entry(q, &dev_queue->qdisc_list, list) { | |
1da177e4 LT |
1030 | if (q_idx < s_q_idx) { |
1031 | q_idx++; | |
1032 | continue; | |
1033 | } | |
1034 | if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
0463d4ae | 1035 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) |
1da177e4 | 1036 | goto done; |
1da177e4 LT |
1037 | q_idx++; |
1038 | } | |
7562f876 PE |
1039 | cont: |
1040 | idx++; | |
1da177e4 LT |
1041 | } |
1042 | ||
1043 | done: | |
1044 | read_unlock(&dev_base_lock); | |
1045 | ||
1046 | cb->args[0] = idx; | |
1047 | cb->args[1] = q_idx; | |
1048 | ||
1049 | return skb->len; | |
1050 | } | |
1051 | ||
1052 | ||
1053 | ||
1054 | /************************************************ | |
1055 | * Traffic classes manipulation. * | |
1056 | ************************************************/ | |
1057 | ||
1058 | ||
1059 | ||
1060 | static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
1061 | { | |
3b1e0a65 | 1062 | struct net *net = sock_net(skb->sk); |
b0e1e646 | 1063 | struct netdev_queue *dev_queue; |
1da177e4 | 1064 | struct tcmsg *tcm = NLMSG_DATA(n); |
1e90474c | 1065 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
1066 | struct net_device *dev; |
1067 | struct Qdisc *q = NULL; | |
20fea08b | 1068 | const struct Qdisc_class_ops *cops; |
1da177e4 LT |
1069 | unsigned long cl = 0; |
1070 | unsigned long new_cl; | |
1071 | u32 pid = tcm->tcm_parent; | |
1072 | u32 clid = tcm->tcm_handle; | |
1073 | u32 qid = TC_H_MAJ(clid); | |
1074 | int err; | |
1075 | ||
b854272b DL |
1076 | if (net != &init_net) |
1077 | return -EINVAL; | |
1078 | ||
881d966b | 1079 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
1080 | return -ENODEV; |
1081 | ||
1e90474c PM |
1082 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); |
1083 | if (err < 0) | |
1084 | return err; | |
1085 | ||
1da177e4 LT |
1086 | /* |
1087 | parent == TC_H_UNSPEC - unspecified parent. | |
1088 | parent == TC_H_ROOT - class is root, which has no parent. | |
1089 | parent == X:0 - parent is root class. | |
1090 | parent == X:Y - parent is a node in hierarchy. | |
1091 | parent == 0:Y - parent is X:Y, where X:0 is qdisc. | |
1092 | ||
1093 | handle == 0:0 - generate handle from kernel pool. | |
1094 | handle == 0:Y - class is X:Y, where X:0 is qdisc. | |
1095 | handle == X:Y - clear. | |
1096 | handle == X:0 - root class. | |
1097 | */ | |
1098 | ||
1099 | /* Step 1. Determine qdisc handle X:0 */ | |
1100 | ||
b0e1e646 | 1101 | dev_queue = &dev->tx_queue; |
1da177e4 LT |
1102 | if (pid != TC_H_ROOT) { |
1103 | u32 qid1 = TC_H_MAJ(pid); | |
1104 | ||
1105 | if (qid && qid1) { | |
1106 | /* If both majors are known, they must be identical. */ | |
1107 | if (qid != qid1) | |
1108 | return -EINVAL; | |
1109 | } else if (qid1) { | |
1110 | qid = qid1; | |
1111 | } else if (qid == 0) | |
b0e1e646 | 1112 | qid = dev_queue->qdisc_sleeping->handle; |
1da177e4 LT |
1113 | |
1114 | /* Now qid is genuine qdisc handle consistent | |
1115 | both with parent and child. | |
1116 | ||
1117 | TC_H_MAJ(pid) still may be unspecified, complete it now. | |
1118 | */ | |
1119 | if (pid) | |
1120 | pid = TC_H_MAKE(qid, pid); | |
1121 | } else { | |
1122 | if (qid == 0) | |
b0e1e646 | 1123 | qid = dev_queue->qdisc_sleeping->handle; |
1da177e4 LT |
1124 | } |
1125 | ||
1126 | /* OK. Locate qdisc */ | |
10297b99 | 1127 | if ((q = qdisc_lookup(dev, qid)) == NULL) |
1da177e4 LT |
1128 | return -ENOENT; |
1129 | ||
1130 | /* An check that it supports classes */ | |
1131 | cops = q->ops->cl_ops; | |
1132 | if (cops == NULL) | |
1133 | return -EINVAL; | |
1134 | ||
1135 | /* Now try to get class */ | |
1136 | if (clid == 0) { | |
1137 | if (pid == TC_H_ROOT) | |
1138 | clid = qid; | |
1139 | } else | |
1140 | clid = TC_H_MAKE(qid, clid); | |
1141 | ||
1142 | if (clid) | |
1143 | cl = cops->get(q, clid); | |
1144 | ||
1145 | if (cl == 0) { | |
1146 | err = -ENOENT; | |
1147 | if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE)) | |
1148 | goto out; | |
1149 | } else { | |
1150 | switch (n->nlmsg_type) { | |
10297b99 | 1151 | case RTM_NEWTCLASS: |
1da177e4 LT |
1152 | err = -EEXIST; |
1153 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1154 | goto out; | |
1155 | break; | |
1156 | case RTM_DELTCLASS: | |
1157 | err = cops->delete(q, cl); | |
1158 | if (err == 0) | |
1159 | tclass_notify(skb, n, q, cl, RTM_DELTCLASS); | |
1160 | goto out; | |
1161 | case RTM_GETTCLASS: | |
1162 | err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS); | |
1163 | goto out; | |
1164 | default: | |
1165 | err = -EINVAL; | |
1166 | goto out; | |
1167 | } | |
1168 | } | |
1169 | ||
1170 | new_cl = cl; | |
1171 | err = cops->change(q, clid, pid, tca, &new_cl); | |
1172 | if (err == 0) | |
1173 | tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS); | |
1174 | ||
1175 | out: | |
1176 | if (cl) | |
1177 | cops->put(q, cl); | |
1178 | ||
1179 | return err; | |
1180 | } | |
1181 | ||
1182 | ||
1183 | static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, | |
1184 | unsigned long cl, | |
e431b8c0 | 1185 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
1186 | { |
1187 | struct tcmsg *tcm; | |
1188 | struct nlmsghdr *nlh; | |
27a884dc | 1189 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 | 1190 | struct gnet_dump d; |
20fea08b | 1191 | const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; |
1da177e4 | 1192 | |
e431b8c0 | 1193 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
1194 | tcm = NLMSG_DATA(nlh); |
1195 | tcm->tcm_family = AF_UNSPEC; | |
5ce2d488 | 1196 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex; |
1da177e4 LT |
1197 | tcm->tcm_parent = q->handle; |
1198 | tcm->tcm_handle = q->handle; | |
1199 | tcm->tcm_info = 0; | |
57e1c487 | 1200 | NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); |
1da177e4 | 1201 | if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) |
1e90474c | 1202 | goto nla_put_failure; |
1da177e4 LT |
1203 | |
1204 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
1205 | TCA_XSTATS, q->stats_lock, &d) < 0) | |
1e90474c | 1206 | goto nla_put_failure; |
1da177e4 LT |
1207 | |
1208 | if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) | |
1e90474c | 1209 | goto nla_put_failure; |
1da177e4 LT |
1210 | |
1211 | if (gnet_stats_finish_copy(&d) < 0) | |
1e90474c | 1212 | goto nla_put_failure; |
1da177e4 | 1213 | |
27a884dc | 1214 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
1215 | return skb->len; |
1216 | ||
1217 | nlmsg_failure: | |
1e90474c | 1218 | nla_put_failure: |
dc5fc579 | 1219 | nlmsg_trim(skb, b); |
1da177e4 LT |
1220 | return -1; |
1221 | } | |
1222 | ||
1223 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1224 | struct Qdisc *q, unsigned long cl, int event) | |
1225 | { | |
1226 | struct sk_buff *skb; | |
1227 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1228 | ||
1229 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1230 | if (!skb) | |
1231 | return -ENOBUFS; | |
1232 | ||
1233 | if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) { | |
1234 | kfree_skb(skb); | |
1235 | return -EINVAL; | |
1236 | } | |
1237 | ||
97c53cac | 1238 | return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
1239 | } |
1240 | ||
1241 | struct qdisc_dump_args | |
1242 | { | |
1243 | struct qdisc_walker w; | |
1244 | struct sk_buff *skb; | |
1245 | struct netlink_callback *cb; | |
1246 | }; | |
1247 | ||
1248 | static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) | |
1249 | { | |
1250 | struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; | |
1251 | ||
1252 | return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid, | |
1253 | a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); | |
1254 | } | |
1255 | ||
1256 | static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) | |
1257 | { | |
3b1e0a65 | 1258 | struct net *net = sock_net(skb->sk); |
b0e1e646 | 1259 | struct netdev_queue *dev_queue; |
1da177e4 LT |
1260 | int t; |
1261 | int s_t; | |
1262 | struct net_device *dev; | |
1263 | struct Qdisc *q; | |
1264 | struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); | |
1265 | struct qdisc_dump_args arg; | |
1266 | ||
b854272b DL |
1267 | if (net != &init_net) |
1268 | return 0; | |
1269 | ||
1da177e4 LT |
1270 | if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) |
1271 | return 0; | |
881d966b | 1272 | if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
1273 | return 0; |
1274 | ||
1275 | s_t = cb->args[0]; | |
1276 | t = 0; | |
1277 | ||
b0e1e646 DM |
1278 | dev_queue = &dev->tx_queue; |
1279 | list_for_each_entry(q, &dev_queue->qdisc_list, list) { | |
1da177e4 LT |
1280 | if (t < s_t || !q->ops->cl_ops || |
1281 | (tcm->tcm_parent && | |
1282 | TC_H_MAJ(tcm->tcm_parent) != q->handle)) { | |
1283 | t++; | |
1284 | continue; | |
1285 | } | |
1286 | if (t > s_t) | |
1287 | memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); | |
1288 | arg.w.fn = qdisc_class_dump; | |
1289 | arg.skb = skb; | |
1290 | arg.cb = cb; | |
1291 | arg.w.stop = 0; | |
1292 | arg.w.skip = cb->args[1]; | |
1293 | arg.w.count = 0; | |
1294 | q->ops->cl_ops->walk(q, &arg.w); | |
1295 | cb->args[1] = arg.w.count; | |
1296 | if (arg.w.stop) | |
1297 | break; | |
1298 | t++; | |
1299 | } | |
1da177e4 LT |
1300 | |
1301 | cb->args[0] = t; | |
1302 | ||
1303 | dev_put(dev); | |
1304 | return skb->len; | |
1305 | } | |
1306 | ||
1307 | /* Main classifier routine: scans classifier chain attached | |
1308 | to this qdisc, (optionally) tests for protocol and asks | |
1309 | specific classifiers. | |
1310 | */ | |
73ca4918 PM |
1311 | int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp, |
1312 | struct tcf_result *res) | |
1313 | { | |
1314 | __be16 protocol = skb->protocol; | |
1315 | int err = 0; | |
1316 | ||
1317 | for (; tp; tp = tp->next) { | |
1318 | if ((tp->protocol == protocol || | |
1319 | tp->protocol == htons(ETH_P_ALL)) && | |
1320 | (err = tp->classify(skb, tp, res)) >= 0) { | |
1321 | #ifdef CONFIG_NET_CLS_ACT | |
1322 | if (err != TC_ACT_RECLASSIFY && skb->tc_verd) | |
1323 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0); | |
1324 | #endif | |
1325 | return err; | |
1326 | } | |
1327 | } | |
1328 | return -1; | |
1329 | } | |
1330 | EXPORT_SYMBOL(tc_classify_compat); | |
1331 | ||
1da177e4 | 1332 | int tc_classify(struct sk_buff *skb, struct tcf_proto *tp, |
73ca4918 | 1333 | struct tcf_result *res) |
1da177e4 LT |
1334 | { |
1335 | int err = 0; | |
73ca4918 | 1336 | __be16 protocol; |
1da177e4 LT |
1337 | #ifdef CONFIG_NET_CLS_ACT |
1338 | struct tcf_proto *otp = tp; | |
1339 | reclassify: | |
1340 | #endif | |
1341 | protocol = skb->protocol; | |
1342 | ||
73ca4918 | 1343 | err = tc_classify_compat(skb, tp, res); |
1da177e4 | 1344 | #ifdef CONFIG_NET_CLS_ACT |
73ca4918 PM |
1345 | if (err == TC_ACT_RECLASSIFY) { |
1346 | u32 verd = G_TC_VERD(skb->tc_verd); | |
1347 | tp = otp; | |
1348 | ||
1349 | if (verd++ >= MAX_REC_LOOP) { | |
1350 | printk("rule prio %u protocol %02x reclassify loop, " | |
1351 | "packet dropped\n", | |
1352 | tp->prio&0xffff, ntohs(tp->protocol)); | |
1353 | return TC_ACT_SHOT; | |
1da177e4 | 1354 | } |
73ca4918 PM |
1355 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd); |
1356 | goto reclassify; | |
1da177e4 | 1357 | } |
73ca4918 PM |
1358 | #endif |
1359 | return err; | |
1da177e4 | 1360 | } |
73ca4918 | 1361 | EXPORT_SYMBOL(tc_classify); |
1da177e4 | 1362 | |
a48b5a61 PM |
1363 | void tcf_destroy(struct tcf_proto *tp) |
1364 | { | |
1365 | tp->ops->destroy(tp); | |
1366 | module_put(tp->ops->owner); | |
1367 | kfree(tp); | |
1368 | } | |
1369 | ||
ff31ab56 | 1370 | void tcf_destroy_chain(struct tcf_proto **fl) |
a48b5a61 PM |
1371 | { |
1372 | struct tcf_proto *tp; | |
1373 | ||
ff31ab56 PM |
1374 | while ((tp = *fl) != NULL) { |
1375 | *fl = tp->next; | |
a48b5a61 PM |
1376 | tcf_destroy(tp); |
1377 | } | |
1378 | } | |
1379 | EXPORT_SYMBOL(tcf_destroy_chain); | |
1380 | ||
1da177e4 LT |
1381 | #ifdef CONFIG_PROC_FS |
1382 | static int psched_show(struct seq_file *seq, void *v) | |
1383 | { | |
3c0cfc13 PM |
1384 | struct timespec ts; |
1385 | ||
1386 | hrtimer_get_res(CLOCK_MONOTONIC, &ts); | |
1da177e4 | 1387 | seq_printf(seq, "%08x %08x %08x %08x\n", |
641b9e0e | 1388 | (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1), |
514bca32 | 1389 | 1000000, |
3c0cfc13 | 1390 | (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts))); |
1da177e4 LT |
1391 | |
1392 | return 0; | |
1393 | } | |
1394 | ||
1395 | static int psched_open(struct inode *inode, struct file *file) | |
1396 | { | |
1397 | return single_open(file, psched_show, PDE(inode)->data); | |
1398 | } | |
1399 | ||
da7071d7 | 1400 | static const struct file_operations psched_fops = { |
1da177e4 LT |
1401 | .owner = THIS_MODULE, |
1402 | .open = psched_open, | |
1403 | .read = seq_read, | |
1404 | .llseek = seq_lseek, | |
1405 | .release = single_release, | |
10297b99 | 1406 | }; |
1da177e4 LT |
1407 | #endif |
1408 | ||
1da177e4 LT |
1409 | static int __init pktsched_init(void) |
1410 | { | |
1da177e4 LT |
1411 | register_qdisc(&pfifo_qdisc_ops); |
1412 | register_qdisc(&bfifo_qdisc_ops); | |
457c4cbc | 1413 | proc_net_fops_create(&init_net, "psched", 0, &psched_fops); |
1da177e4 | 1414 | |
be577ddc TG |
1415 | rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL); |
1416 | rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL); | |
1417 | rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc); | |
1418 | rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL); | |
1419 | rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL); | |
1420 | rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass); | |
1421 | ||
1da177e4 LT |
1422 | return 0; |
1423 | } | |
1424 | ||
1425 | subsys_initcall(pktsched_init); |