]>
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 | ||
ead81cc5 | 186 | struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) |
1da177e4 | 187 | { |
30723673 DM |
188 | unsigned int i; |
189 | ||
190 | for (i = 0; i < dev->num_tx_queues; i++) { | |
191 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
192 | struct Qdisc *q, *txq_root = txq->qdisc; | |
1da177e4 | 193 | |
30723673 DM |
194 | if (!(txq_root->flags & TCQ_F_BUILTIN) && |
195 | txq_root->handle == handle) | |
196 | return txq_root; | |
197 | ||
198 | list_for_each_entry(q, &txq_root->list, list) { | |
199 | if (q->handle == handle) | |
200 | return q; | |
201 | } | |
1da177e4 | 202 | } |
1da177e4 LT |
203 | return NULL; |
204 | } | |
205 | ||
206 | static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) | |
207 | { | |
208 | unsigned long cl; | |
209 | struct Qdisc *leaf; | |
20fea08b | 210 | const struct Qdisc_class_ops *cops = p->ops->cl_ops; |
1da177e4 LT |
211 | |
212 | if (cops == NULL) | |
213 | return NULL; | |
214 | cl = cops->get(p, classid); | |
215 | ||
216 | if (cl == 0) | |
217 | return NULL; | |
218 | leaf = cops->leaf(p, cl); | |
219 | cops->put(p, cl); | |
220 | return leaf; | |
221 | } | |
222 | ||
223 | /* Find queueing discipline by name */ | |
224 | ||
1e90474c | 225 | static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) |
1da177e4 LT |
226 | { |
227 | struct Qdisc_ops *q = NULL; | |
228 | ||
229 | if (kind) { | |
230 | read_lock(&qdisc_mod_lock); | |
231 | for (q = qdisc_base; q; q = q->next) { | |
1e90474c | 232 | if (nla_strcmp(kind, q->id) == 0) { |
1da177e4 LT |
233 | if (!try_module_get(q->owner)) |
234 | q = NULL; | |
235 | break; | |
236 | } | |
237 | } | |
238 | read_unlock(&qdisc_mod_lock); | |
239 | } | |
240 | return q; | |
241 | } | |
242 | ||
243 | static struct qdisc_rate_table *qdisc_rtab_list; | |
244 | ||
1e90474c | 245 | struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab) |
1da177e4 LT |
246 | { |
247 | struct qdisc_rate_table *rtab; | |
248 | ||
249 | for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { | |
250 | if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) { | |
251 | rtab->refcnt++; | |
252 | return rtab; | |
253 | } | |
254 | } | |
255 | ||
5feb5e1a PM |
256 | if (tab == NULL || r->rate == 0 || r->cell_log == 0 || |
257 | nla_len(tab) != TC_RTAB_SIZE) | |
1da177e4 LT |
258 | return NULL; |
259 | ||
260 | rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); | |
261 | if (rtab) { | |
262 | rtab->rate = *r; | |
263 | rtab->refcnt = 1; | |
1e90474c | 264 | memcpy(rtab->data, nla_data(tab), 1024); |
1da177e4 LT |
265 | rtab->next = qdisc_rtab_list; |
266 | qdisc_rtab_list = rtab; | |
267 | } | |
268 | return rtab; | |
269 | } | |
62e3ba1b | 270 | EXPORT_SYMBOL(qdisc_get_rtab); |
1da177e4 LT |
271 | |
272 | void qdisc_put_rtab(struct qdisc_rate_table *tab) | |
273 | { | |
274 | struct qdisc_rate_table *rtab, **rtabp; | |
275 | ||
276 | if (!tab || --tab->refcnt) | |
277 | return; | |
278 | ||
279 | for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) { | |
280 | if (rtab == tab) { | |
281 | *rtabp = rtab->next; | |
282 | kfree(rtab); | |
283 | return; | |
284 | } | |
285 | } | |
286 | } | |
62e3ba1b | 287 | EXPORT_SYMBOL(qdisc_put_rtab); |
1da177e4 | 288 | |
4179477f PM |
289 | static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) |
290 | { | |
291 | struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, | |
292 | timer); | |
293 | ||
294 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
11274e5a | 295 | smp_wmb(); |
37437bb2 | 296 | __netif_schedule(wd->qdisc); |
1936502d | 297 | |
4179477f PM |
298 | return HRTIMER_NORESTART; |
299 | } | |
300 | ||
301 | void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) | |
302 | { | |
303 | hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
304 | wd->timer.function = qdisc_watchdog; | |
305 | wd->qdisc = qdisc; | |
306 | } | |
307 | EXPORT_SYMBOL(qdisc_watchdog_init); | |
308 | ||
309 | void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires) | |
310 | { | |
311 | ktime_t time; | |
312 | ||
313 | wd->qdisc->flags |= TCQ_F_THROTTLED; | |
314 | time = ktime_set(0, 0); | |
315 | time = ktime_add_ns(time, PSCHED_US2NS(expires)); | |
316 | hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS); | |
317 | } | |
318 | EXPORT_SYMBOL(qdisc_watchdog_schedule); | |
319 | ||
320 | void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) | |
321 | { | |
322 | hrtimer_cancel(&wd->timer); | |
323 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
324 | } | |
325 | EXPORT_SYMBOL(qdisc_watchdog_cancel); | |
1da177e4 | 326 | |
6fe1c7a5 PM |
327 | struct hlist_head *qdisc_class_hash_alloc(unsigned int n) |
328 | { | |
329 | unsigned int size = n * sizeof(struct hlist_head), i; | |
330 | struct hlist_head *h; | |
331 | ||
332 | if (size <= PAGE_SIZE) | |
333 | h = kmalloc(size, GFP_KERNEL); | |
334 | else | |
335 | h = (struct hlist_head *) | |
336 | __get_free_pages(GFP_KERNEL, get_order(size)); | |
337 | ||
338 | if (h != NULL) { | |
339 | for (i = 0; i < n; i++) | |
340 | INIT_HLIST_HEAD(&h[i]); | |
341 | } | |
342 | return h; | |
343 | } | |
344 | ||
345 | static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) | |
346 | { | |
347 | unsigned int size = n * sizeof(struct hlist_head); | |
348 | ||
349 | if (size <= PAGE_SIZE) | |
350 | kfree(h); | |
351 | else | |
352 | free_pages((unsigned long)h, get_order(size)); | |
353 | } | |
354 | ||
355 | void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) | |
356 | { | |
357 | struct Qdisc_class_common *cl; | |
358 | struct hlist_node *n, *next; | |
359 | struct hlist_head *nhash, *ohash; | |
360 | unsigned int nsize, nmask, osize; | |
361 | unsigned int i, h; | |
362 | ||
363 | /* Rehash when load factor exceeds 0.75 */ | |
364 | if (clhash->hashelems * 4 <= clhash->hashsize * 3) | |
365 | return; | |
366 | nsize = clhash->hashsize * 2; | |
367 | nmask = nsize - 1; | |
368 | nhash = qdisc_class_hash_alloc(nsize); | |
369 | if (nhash == NULL) | |
370 | return; | |
371 | ||
372 | ohash = clhash->hash; | |
373 | osize = clhash->hashsize; | |
374 | ||
375 | sch_tree_lock(sch); | |
376 | for (i = 0; i < osize; i++) { | |
377 | hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) { | |
378 | h = qdisc_class_hash(cl->classid, nmask); | |
379 | hlist_add_head(&cl->hnode, &nhash[h]); | |
380 | } | |
381 | } | |
382 | clhash->hash = nhash; | |
383 | clhash->hashsize = nsize; | |
384 | clhash->hashmask = nmask; | |
385 | sch_tree_unlock(sch); | |
386 | ||
387 | qdisc_class_hash_free(ohash, osize); | |
388 | } | |
389 | EXPORT_SYMBOL(qdisc_class_hash_grow); | |
390 | ||
391 | int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) | |
392 | { | |
393 | unsigned int size = 4; | |
394 | ||
395 | clhash->hash = qdisc_class_hash_alloc(size); | |
396 | if (clhash->hash == NULL) | |
397 | return -ENOMEM; | |
398 | clhash->hashsize = size; | |
399 | clhash->hashmask = size - 1; | |
400 | clhash->hashelems = 0; | |
401 | return 0; | |
402 | } | |
403 | EXPORT_SYMBOL(qdisc_class_hash_init); | |
404 | ||
405 | void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) | |
406 | { | |
407 | qdisc_class_hash_free(clhash->hash, clhash->hashsize); | |
408 | } | |
409 | EXPORT_SYMBOL(qdisc_class_hash_destroy); | |
410 | ||
411 | void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, | |
412 | struct Qdisc_class_common *cl) | |
413 | { | |
414 | unsigned int h; | |
415 | ||
416 | INIT_HLIST_NODE(&cl->hnode); | |
417 | h = qdisc_class_hash(cl->classid, clhash->hashmask); | |
418 | hlist_add_head(&cl->hnode, &clhash->hash[h]); | |
419 | clhash->hashelems++; | |
420 | } | |
421 | EXPORT_SYMBOL(qdisc_class_hash_insert); | |
422 | ||
423 | void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, | |
424 | struct Qdisc_class_common *cl) | |
425 | { | |
426 | hlist_del(&cl->hnode); | |
427 | clhash->hashelems--; | |
428 | } | |
429 | EXPORT_SYMBOL(qdisc_class_hash_remove); | |
430 | ||
1da177e4 LT |
431 | /* Allocate an unique handle from space managed by kernel */ |
432 | ||
433 | static u32 qdisc_alloc_handle(struct net_device *dev) | |
434 | { | |
435 | int i = 0x10000; | |
436 | static u32 autohandle = TC_H_MAKE(0x80000000U, 0); | |
437 | ||
438 | do { | |
439 | autohandle += TC_H_MAKE(0x10000U, 0); | |
440 | if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) | |
441 | autohandle = TC_H_MAKE(0x80000000U, 0); | |
442 | } while (qdisc_lookup(dev, autohandle) && --i > 0); | |
443 | ||
444 | return i>0 ? autohandle : 0; | |
445 | } | |
446 | ||
99194cff | 447 | /* Attach toplevel qdisc to device queue. */ |
1da177e4 | 448 | |
99194cff DM |
449 | static struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, |
450 | struct Qdisc *qdisc) | |
1da177e4 | 451 | { |
53049978 | 452 | spinlock_t *root_lock; |
1da177e4 | 453 | struct Qdisc *oqdisc; |
53049978 | 454 | int ingress; |
1da177e4 | 455 | |
53049978 DM |
456 | ingress = 0; |
457 | if (qdisc && qdisc->flags&TCQ_F_INGRESS) | |
458 | ingress = 1; | |
459 | ||
460 | if (ingress) { | |
816f3258 | 461 | oqdisc = dev_queue->qdisc; |
53049978 | 462 | } else { |
53049978 DM |
463 | oqdisc = dev_queue->qdisc_sleeping; |
464 | } | |
465 | ||
466 | root_lock = qdisc_root_lock(oqdisc); | |
467 | spin_lock_bh(root_lock); | |
468 | ||
469 | if (ingress) { | |
1da177e4 LT |
470 | /* Prune old scheduler */ |
471 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) { | |
472 | /* delete */ | |
473 | qdisc_reset(oqdisc); | |
816f3258 | 474 | dev_queue->qdisc = NULL; |
1da177e4 | 475 | } else { /* new */ |
816f3258 | 476 | dev_queue->qdisc = qdisc; |
1da177e4 LT |
477 | } |
478 | ||
479 | } else { | |
1da177e4 LT |
480 | /* Prune old scheduler */ |
481 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
482 | qdisc_reset(oqdisc); | |
483 | ||
484 | /* ... and graft new one */ | |
485 | if (qdisc == NULL) | |
486 | qdisc = &noop_qdisc; | |
b0e1e646 DM |
487 | dev_queue->qdisc_sleeping = qdisc; |
488 | dev_queue->qdisc = &noop_qdisc; | |
1da177e4 LT |
489 | } |
490 | ||
53049978 | 491 | spin_unlock_bh(root_lock); |
1da177e4 | 492 | |
1da177e4 LT |
493 | return oqdisc; |
494 | } | |
495 | ||
43effa1e PM |
496 | void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n) |
497 | { | |
20fea08b | 498 | const struct Qdisc_class_ops *cops; |
43effa1e PM |
499 | unsigned long cl; |
500 | u32 parentid; | |
501 | ||
502 | if (n == 0) | |
503 | return; | |
504 | while ((parentid = sch->parent)) { | |
066a3b5b JP |
505 | if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) |
506 | return; | |
507 | ||
5ce2d488 | 508 | sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); |
ffc8fefa PM |
509 | if (sch == NULL) { |
510 | WARN_ON(parentid != TC_H_ROOT); | |
511 | return; | |
512 | } | |
43effa1e PM |
513 | cops = sch->ops->cl_ops; |
514 | if (cops->qlen_notify) { | |
515 | cl = cops->get(sch, parentid); | |
516 | cops->qlen_notify(sch, cl); | |
517 | cops->put(sch, cl); | |
518 | } | |
519 | sch->q.qlen -= n; | |
520 | } | |
521 | } | |
522 | EXPORT_SYMBOL(qdisc_tree_decrease_qlen); | |
1da177e4 | 523 | |
99194cff DM |
524 | static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid, |
525 | struct Qdisc *old, struct Qdisc *new) | |
526 | { | |
527 | if (new || old) | |
528 | qdisc_notify(skb, n, clid, old, new); | |
1da177e4 | 529 | |
99194cff DM |
530 | if (old) { |
531 | spin_lock_bh(&old->q.lock); | |
532 | qdisc_destroy(old); | |
533 | spin_unlock_bh(&old->q.lock); | |
534 | } | |
535 | } | |
536 | ||
537 | /* Graft qdisc "new" to class "classid" of qdisc "parent" or | |
538 | * to device "dev". | |
539 | * | |
540 | * When appropriate send a netlink notification using 'skb' | |
541 | * and "n". | |
542 | * | |
543 | * On success, destroy old qdisc. | |
1da177e4 LT |
544 | */ |
545 | ||
546 | static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, | |
99194cff DM |
547 | struct sk_buff *skb, struct nlmsghdr *n, u32 classid, |
548 | struct Qdisc *new, struct Qdisc *old) | |
1da177e4 | 549 | { |
99194cff | 550 | struct Qdisc *q = old; |
1da177e4 | 551 | int err = 0; |
1da177e4 | 552 | |
10297b99 | 553 | if (parent == NULL) { |
99194cff DM |
554 | unsigned int i, num_q, ingress; |
555 | ||
556 | ingress = 0; | |
557 | num_q = dev->num_tx_queues; | |
558 | if (q && q->flags & TCQ_F_INGRESS) { | |
559 | num_q = 1; | |
560 | ingress = 1; | |
561 | } | |
562 | ||
563 | if (dev->flags & IFF_UP) | |
564 | dev_deactivate(dev); | |
565 | ||
566 | for (i = 0; i < num_q; i++) { | |
567 | struct netdev_queue *dev_queue = &dev->rx_queue; | |
568 | ||
569 | if (!ingress) | |
570 | dev_queue = netdev_get_tx_queue(dev, i); | |
571 | ||
572 | if (ingress) { | |
573 | old = dev_graft_qdisc(dev_queue, q); | |
574 | } else { | |
575 | old = dev_graft_qdisc(dev_queue, new); | |
576 | if (new && i > 0) | |
577 | atomic_inc(&new->refcnt); | |
578 | } | |
579 | notify_and_destroy(skb, n, classid, old, new); | |
1da177e4 | 580 | } |
99194cff DM |
581 | |
582 | if (dev->flags & IFF_UP) | |
583 | dev_activate(dev); | |
1da177e4 | 584 | } else { |
20fea08b | 585 | const struct Qdisc_class_ops *cops = parent->ops->cl_ops; |
1da177e4 LT |
586 | |
587 | err = -EINVAL; | |
588 | ||
589 | if (cops) { | |
590 | unsigned long cl = cops->get(parent, classid); | |
591 | if (cl) { | |
99194cff | 592 | err = cops->graft(parent, cl, new, &old); |
1da177e4 LT |
593 | cops->put(parent, cl); |
594 | } | |
595 | } | |
99194cff DM |
596 | if (!err) |
597 | notify_and_destroy(skb, n, classid, old, new); | |
1da177e4 LT |
598 | } |
599 | return err; | |
600 | } | |
601 | ||
602 | /* | |
603 | Allocate and initialize new qdisc. | |
604 | ||
605 | Parameters are passed via opt. | |
606 | */ | |
607 | ||
608 | static struct Qdisc * | |
bb949fbd DM |
609 | qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue, |
610 | u32 parent, u32 handle, struct nlattr **tca, int *errp) | |
1da177e4 LT |
611 | { |
612 | int err; | |
1e90474c | 613 | struct nlattr *kind = tca[TCA_KIND]; |
1da177e4 LT |
614 | struct Qdisc *sch; |
615 | struct Qdisc_ops *ops; | |
1da177e4 LT |
616 | |
617 | ops = qdisc_lookup_ops(kind); | |
618 | #ifdef CONFIG_KMOD | |
619 | if (ops == NULL && kind != NULL) { | |
620 | char name[IFNAMSIZ]; | |
1e90474c | 621 | if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { |
1da177e4 LT |
622 | /* We dropped the RTNL semaphore in order to |
623 | * perform the module load. So, even if we | |
624 | * succeeded in loading the module we have to | |
625 | * tell the caller to replay the request. We | |
626 | * indicate this using -EAGAIN. | |
627 | * We replay the request because the device may | |
628 | * go away in the mean time. | |
629 | */ | |
630 | rtnl_unlock(); | |
631 | request_module("sch_%s", name); | |
632 | rtnl_lock(); | |
633 | ops = qdisc_lookup_ops(kind); | |
634 | if (ops != NULL) { | |
635 | /* We will try again qdisc_lookup_ops, | |
636 | * so don't keep a reference. | |
637 | */ | |
638 | module_put(ops->owner); | |
639 | err = -EAGAIN; | |
640 | goto err_out; | |
641 | } | |
642 | } | |
643 | } | |
644 | #endif | |
645 | ||
b9e2cc0f | 646 | err = -ENOENT; |
1da177e4 LT |
647 | if (ops == NULL) |
648 | goto err_out; | |
649 | ||
5ce2d488 | 650 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
651 | if (IS_ERR(sch)) { |
652 | err = PTR_ERR(sch); | |
1da177e4 | 653 | goto err_out2; |
3d54b82f | 654 | } |
1da177e4 | 655 | |
ffc8fefa PM |
656 | sch->parent = parent; |
657 | ||
3d54b82f | 658 | if (handle == TC_H_INGRESS) { |
1da177e4 | 659 | sch->flags |= TCQ_F_INGRESS; |
3d54b82f | 660 | handle = TC_H_MAKE(TC_H_INGRESS, 0); |
fd44de7c | 661 | } else { |
fd44de7c PM |
662 | if (handle == 0) { |
663 | handle = qdisc_alloc_handle(dev); | |
664 | err = -ENOMEM; | |
665 | if (handle == 0) | |
666 | goto err_out3; | |
667 | } | |
1da177e4 LT |
668 | } |
669 | ||
3d54b82f | 670 | sch->handle = handle; |
1da177e4 | 671 | |
1e90474c PM |
672 | if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { |
673 | if (tca[TCA_RATE]) { | |
023e09a7 | 674 | err = gen_new_estimator(&sch->bstats, &sch->rate_est, |
7698b4fc | 675 | qdisc_root_lock(sch), |
1e90474c | 676 | tca[TCA_RATE]); |
023e09a7 TG |
677 | if (err) { |
678 | /* | |
679 | * Any broken qdiscs that would require | |
680 | * a ops->reset() here? The qdisc was never | |
681 | * in action so it shouldn't be necessary. | |
682 | */ | |
683 | if (ops->destroy) | |
684 | ops->destroy(sch); | |
685 | goto err_out3; | |
686 | } | |
687 | } | |
30723673 DM |
688 | if (parent) |
689 | list_add_tail(&sch->list, &dev_queue->qdisc->list); | |
1da177e4 | 690 | |
1da177e4 LT |
691 | return sch; |
692 | } | |
693 | err_out3: | |
694 | dev_put(dev); | |
3d54b82f | 695 | kfree((char *) sch - sch->padded); |
1da177e4 LT |
696 | err_out2: |
697 | module_put(ops->owner); | |
698 | err_out: | |
699 | *errp = err; | |
1da177e4 LT |
700 | return NULL; |
701 | } | |
702 | ||
1e90474c | 703 | static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) |
1da177e4 | 704 | { |
1e90474c | 705 | if (tca[TCA_OPTIONS]) { |
1da177e4 LT |
706 | int err; |
707 | ||
708 | if (sch->ops->change == NULL) | |
709 | return -EINVAL; | |
1e90474c | 710 | err = sch->ops->change(sch, tca[TCA_OPTIONS]); |
1da177e4 LT |
711 | if (err) |
712 | return err; | |
713 | } | |
1e90474c | 714 | if (tca[TCA_RATE]) |
1da177e4 | 715 | gen_replace_estimator(&sch->bstats, &sch->rate_est, |
7698b4fc | 716 | qdisc_root_lock(sch), tca[TCA_RATE]); |
1da177e4 LT |
717 | return 0; |
718 | } | |
719 | ||
720 | struct check_loop_arg | |
721 | { | |
722 | struct qdisc_walker w; | |
723 | struct Qdisc *p; | |
724 | int depth; | |
725 | }; | |
726 | ||
727 | static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); | |
728 | ||
729 | static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) | |
730 | { | |
731 | struct check_loop_arg arg; | |
732 | ||
733 | if (q->ops->cl_ops == NULL) | |
734 | return 0; | |
735 | ||
736 | arg.w.stop = arg.w.skip = arg.w.count = 0; | |
737 | arg.w.fn = check_loop_fn; | |
738 | arg.depth = depth; | |
739 | arg.p = p; | |
740 | q->ops->cl_ops->walk(q, &arg.w); | |
741 | return arg.w.stop ? -ELOOP : 0; | |
742 | } | |
743 | ||
744 | static int | |
745 | check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) | |
746 | { | |
747 | struct Qdisc *leaf; | |
20fea08b | 748 | const struct Qdisc_class_ops *cops = q->ops->cl_ops; |
1da177e4 LT |
749 | struct check_loop_arg *arg = (struct check_loop_arg *)w; |
750 | ||
751 | leaf = cops->leaf(q, cl); | |
752 | if (leaf) { | |
753 | if (leaf == arg->p || arg->depth > 7) | |
754 | return -ELOOP; | |
755 | return check_loop(leaf, arg->p, arg->depth + 1); | |
756 | } | |
757 | return 0; | |
758 | } | |
759 | ||
760 | /* | |
761 | * Delete/get qdisc. | |
762 | */ | |
763 | ||
764 | static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
765 | { | |
3b1e0a65 | 766 | struct net *net = sock_net(skb->sk); |
1da177e4 | 767 | struct tcmsg *tcm = NLMSG_DATA(n); |
1e90474c | 768 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
769 | struct net_device *dev; |
770 | u32 clid = tcm->tcm_parent; | |
771 | struct Qdisc *q = NULL; | |
772 | struct Qdisc *p = NULL; | |
773 | int err; | |
774 | ||
b854272b DL |
775 | if (net != &init_net) |
776 | return -EINVAL; | |
777 | ||
881d966b | 778 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
779 | return -ENODEV; |
780 | ||
1e90474c PM |
781 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); |
782 | if (err < 0) | |
783 | return err; | |
784 | ||
1da177e4 LT |
785 | if (clid) { |
786 | if (clid != TC_H_ROOT) { | |
787 | if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { | |
788 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
789 | return -ENOENT; | |
790 | q = qdisc_leaf(p, clid); | |
791 | } else { /* ingress */ | |
816f3258 | 792 | q = dev->rx_queue.qdisc; |
10297b99 | 793 | } |
1da177e4 | 794 | } else { |
e8a0464c DM |
795 | struct netdev_queue *dev_queue; |
796 | dev_queue = netdev_get_tx_queue(dev, 0); | |
b0e1e646 | 797 | q = dev_queue->qdisc_sleeping; |
1da177e4 LT |
798 | } |
799 | if (!q) | |
800 | return -ENOENT; | |
801 | ||
802 | if (tcm->tcm_handle && q->handle != tcm->tcm_handle) | |
803 | return -EINVAL; | |
804 | } else { | |
805 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
806 | return -ENOENT; | |
807 | } | |
808 | ||
1e90474c | 809 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
810 | return -EINVAL; |
811 | ||
812 | if (n->nlmsg_type == RTM_DELQDISC) { | |
813 | if (!clid) | |
814 | return -EINVAL; | |
815 | if (q->handle == 0) | |
816 | return -ENOENT; | |
99194cff | 817 | if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0) |
1da177e4 | 818 | return err; |
1da177e4 LT |
819 | } else { |
820 | qdisc_notify(skb, n, clid, NULL, q); | |
821 | } | |
822 | return 0; | |
823 | } | |
824 | ||
825 | /* | |
826 | Create/change qdisc. | |
827 | */ | |
828 | ||
829 | static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
830 | { | |
3b1e0a65 | 831 | struct net *net = sock_net(skb->sk); |
1da177e4 | 832 | struct tcmsg *tcm; |
1e90474c | 833 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
834 | struct net_device *dev; |
835 | u32 clid; | |
836 | struct Qdisc *q, *p; | |
837 | int err; | |
838 | ||
b854272b DL |
839 | if (net != &init_net) |
840 | return -EINVAL; | |
841 | ||
1da177e4 LT |
842 | replay: |
843 | /* Reinit, just in case something touches this. */ | |
844 | tcm = NLMSG_DATA(n); | |
1da177e4 LT |
845 | clid = tcm->tcm_parent; |
846 | q = p = NULL; | |
847 | ||
881d966b | 848 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
849 | return -ENODEV; |
850 | ||
1e90474c PM |
851 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); |
852 | if (err < 0) | |
853 | return err; | |
854 | ||
1da177e4 LT |
855 | if (clid) { |
856 | if (clid != TC_H_ROOT) { | |
857 | if (clid != TC_H_INGRESS) { | |
858 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
859 | return -ENOENT; | |
860 | q = qdisc_leaf(p, clid); | |
861 | } else { /*ingress */ | |
816f3258 | 862 | q = dev->rx_queue.qdisc; |
1da177e4 LT |
863 | } |
864 | } else { | |
e8a0464c DM |
865 | struct netdev_queue *dev_queue; |
866 | dev_queue = netdev_get_tx_queue(dev, 0); | |
b0e1e646 | 867 | q = dev_queue->qdisc_sleeping; |
1da177e4 LT |
868 | } |
869 | ||
870 | /* It may be default qdisc, ignore it */ | |
871 | if (q && q->handle == 0) | |
872 | q = NULL; | |
873 | ||
874 | if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { | |
875 | if (tcm->tcm_handle) { | |
876 | if (q && !(n->nlmsg_flags&NLM_F_REPLACE)) | |
877 | return -EEXIST; | |
878 | if (TC_H_MIN(tcm->tcm_handle)) | |
879 | return -EINVAL; | |
880 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
881 | goto create_n_graft; | |
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 | if (q == p || | |
887 | (p && check_loop(q, p, 0))) | |
888 | return -ELOOP; | |
889 | atomic_inc(&q->refcnt); | |
890 | goto graft; | |
891 | } else { | |
892 | if (q == NULL) | |
893 | goto create_n_graft; | |
894 | ||
895 | /* This magic test requires explanation. | |
896 | * | |
897 | * We know, that some child q is already | |
898 | * attached to this parent and have choice: | |
899 | * either to change it or to create/graft new one. | |
900 | * | |
901 | * 1. We are allowed to create/graft only | |
902 | * if CREATE and REPLACE flags are set. | |
903 | * | |
904 | * 2. If EXCL is set, requestor wanted to say, | |
905 | * that qdisc tcm_handle is not expected | |
906 | * to exist, so that we choose create/graft too. | |
907 | * | |
908 | * 3. The last case is when no flags are set. | |
909 | * Alas, it is sort of hole in API, we | |
910 | * cannot decide what to do unambiguously. | |
911 | * For now we select create/graft, if | |
912 | * user gave KIND, which does not match existing. | |
913 | */ | |
914 | if ((n->nlmsg_flags&NLM_F_CREATE) && | |
915 | (n->nlmsg_flags&NLM_F_REPLACE) && | |
916 | ((n->nlmsg_flags&NLM_F_EXCL) || | |
1e90474c PM |
917 | (tca[TCA_KIND] && |
918 | nla_strcmp(tca[TCA_KIND], q->ops->id)))) | |
1da177e4 LT |
919 | goto create_n_graft; |
920 | } | |
921 | } | |
922 | } else { | |
923 | if (!tcm->tcm_handle) | |
924 | return -EINVAL; | |
925 | q = qdisc_lookup(dev, tcm->tcm_handle); | |
926 | } | |
927 | ||
928 | /* Change qdisc parameters */ | |
929 | if (q == NULL) | |
930 | return -ENOENT; | |
931 | if (n->nlmsg_flags&NLM_F_EXCL) | |
932 | return -EEXIST; | |
1e90474c | 933 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
934 | return -EINVAL; |
935 | err = qdisc_change(q, tca); | |
936 | if (err == 0) | |
937 | qdisc_notify(skb, n, clid, NULL, q); | |
938 | return err; | |
939 | ||
940 | create_n_graft: | |
941 | if (!(n->nlmsg_flags&NLM_F_CREATE)) | |
942 | return -ENOENT; | |
943 | if (clid == TC_H_INGRESS) | |
bb949fbd DM |
944 | q = qdisc_create(dev, &dev->rx_queue, |
945 | tcm->tcm_parent, tcm->tcm_parent, | |
ffc8fefa | 946 | tca, &err); |
10297b99 | 947 | else |
e8a0464c | 948 | q = qdisc_create(dev, netdev_get_tx_queue(dev, 0), |
bb949fbd | 949 | tcm->tcm_parent, tcm->tcm_handle, |
ffc8fefa | 950 | tca, &err); |
1da177e4 LT |
951 | if (q == NULL) { |
952 | if (err == -EAGAIN) | |
953 | goto replay; | |
954 | return err; | |
955 | } | |
956 | ||
957 | graft: | |
958 | if (1) { | |
53049978 DM |
959 | spinlock_t *root_lock; |
960 | ||
99194cff | 961 | err = qdisc_graft(dev, p, skb, n, clid, q, NULL); |
1da177e4 LT |
962 | if (err) { |
963 | if (q) { | |
53049978 DM |
964 | root_lock = qdisc_root_lock(q); |
965 | spin_lock_bh(root_lock); | |
1da177e4 | 966 | qdisc_destroy(q); |
53049978 | 967 | spin_unlock_bh(root_lock); |
1da177e4 LT |
968 | } |
969 | return err; | |
970 | } | |
1da177e4 LT |
971 | } |
972 | return 0; | |
973 | } | |
974 | ||
975 | static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, | |
e431b8c0 | 976 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
977 | { |
978 | struct tcmsg *tcm; | |
979 | struct nlmsghdr *nlh; | |
27a884dc | 980 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
981 | struct gnet_dump d; |
982 | ||
e431b8c0 | 983 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
984 | tcm = NLMSG_DATA(nlh); |
985 | tcm->tcm_family = AF_UNSPEC; | |
9ef1d4c7 PM |
986 | tcm->tcm__pad1 = 0; |
987 | tcm->tcm__pad2 = 0; | |
5ce2d488 | 988 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex; |
1da177e4 LT |
989 | tcm->tcm_parent = clid; |
990 | tcm->tcm_handle = q->handle; | |
991 | tcm->tcm_info = atomic_read(&q->refcnt); | |
57e1c487 | 992 | NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); |
1da177e4 | 993 | if (q->ops->dump && q->ops->dump(q, skb) < 0) |
1e90474c | 994 | goto nla_put_failure; |
1da177e4 LT |
995 | q->qstats.qlen = q->q.qlen; |
996 | ||
997 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
7698b4fc | 998 | TCA_XSTATS, qdisc_root_lock(q), &d) < 0) |
1e90474c | 999 | goto nla_put_failure; |
1da177e4 LT |
1000 | |
1001 | if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) | |
1e90474c | 1002 | goto nla_put_failure; |
1da177e4 LT |
1003 | |
1004 | if (gnet_stats_copy_basic(&d, &q->bstats) < 0 || | |
1da177e4 | 1005 | gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || |
1da177e4 | 1006 | gnet_stats_copy_queue(&d, &q->qstats) < 0) |
1e90474c | 1007 | goto nla_put_failure; |
10297b99 | 1008 | |
1da177e4 | 1009 | if (gnet_stats_finish_copy(&d) < 0) |
1e90474c | 1010 | goto nla_put_failure; |
10297b99 | 1011 | |
27a884dc | 1012 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
1013 | return skb->len; |
1014 | ||
1015 | nlmsg_failure: | |
1e90474c | 1016 | nla_put_failure: |
dc5fc579 | 1017 | nlmsg_trim(skb, b); |
1da177e4 LT |
1018 | return -1; |
1019 | } | |
1020 | ||
1021 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1022 | u32 clid, struct Qdisc *old, struct Qdisc *new) | |
1023 | { | |
1024 | struct sk_buff *skb; | |
1025 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1026 | ||
1027 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1028 | if (!skb) | |
1029 | return -ENOBUFS; | |
1030 | ||
1031 | if (old && old->handle) { | |
1032 | if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0) | |
1033 | goto err_out; | |
1034 | } | |
1035 | if (new) { | |
1036 | if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) | |
1037 | goto err_out; | |
1038 | } | |
1039 | ||
1040 | if (skb->len) | |
97c53cac | 1041 | return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
1042 | |
1043 | err_out: | |
1044 | kfree_skb(skb); | |
1045 | return -EINVAL; | |
1046 | } | |
1047 | ||
30723673 DM |
1048 | static bool tc_qdisc_dump_ignore(struct Qdisc *q) |
1049 | { | |
1050 | return (q->flags & TCQ_F_BUILTIN) ? true : false; | |
1051 | } | |
1052 | ||
1053 | static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, | |
1054 | struct netlink_callback *cb, | |
1055 | int *q_idx_p, int s_q_idx) | |
1056 | { | |
1057 | int ret = 0, q_idx = *q_idx_p; | |
1058 | struct Qdisc *q; | |
1059 | ||
1060 | if (!root) | |
1061 | return 0; | |
1062 | ||
1063 | q = root; | |
1064 | if (q_idx < s_q_idx) { | |
1065 | q_idx++; | |
1066 | } else { | |
1067 | if (!tc_qdisc_dump_ignore(q) && | |
1068 | tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
1069 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) | |
1070 | goto done; | |
1071 | q_idx++; | |
1072 | } | |
1073 | list_for_each_entry(q, &root->list, list) { | |
1074 | if (q_idx < s_q_idx) { | |
1075 | q_idx++; | |
1076 | continue; | |
1077 | } | |
1078 | if (!tc_qdisc_dump_ignore(q) && | |
1079 | tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
1080 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) | |
1081 | goto done; | |
1082 | q_idx++; | |
1083 | } | |
1084 | ||
1085 | out: | |
1086 | *q_idx_p = q_idx; | |
1087 | return ret; | |
1088 | done: | |
1089 | ret = -1; | |
1090 | goto out; | |
1091 | } | |
1092 | ||
1da177e4 LT |
1093 | static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) |
1094 | { | |
3b1e0a65 | 1095 | struct net *net = sock_net(skb->sk); |
1da177e4 LT |
1096 | int idx, q_idx; |
1097 | int s_idx, s_q_idx; | |
1098 | struct net_device *dev; | |
1da177e4 | 1099 | |
b854272b DL |
1100 | if (net != &init_net) |
1101 | return 0; | |
1102 | ||
1da177e4 LT |
1103 | s_idx = cb->args[0]; |
1104 | s_q_idx = q_idx = cb->args[1]; | |
1105 | read_lock(&dev_base_lock); | |
7562f876 | 1106 | idx = 0; |
881d966b | 1107 | for_each_netdev(&init_net, dev) { |
30723673 DM |
1108 | struct netdev_queue *dev_queue; |
1109 | ||
1da177e4 | 1110 | if (idx < s_idx) |
7562f876 | 1111 | goto cont; |
1da177e4 LT |
1112 | if (idx > s_idx) |
1113 | s_q_idx = 0; | |
1da177e4 | 1114 | q_idx = 0; |
30723673 DM |
1115 | |
1116 | dev_queue = netdev_get_tx_queue(dev, 0); | |
1117 | if (tc_dump_qdisc_root(dev_queue->qdisc, skb, cb, &q_idx, s_q_idx) < 0) | |
1118 | goto done; | |
1119 | ||
1120 | dev_queue = &dev->rx_queue; | |
1121 | if (tc_dump_qdisc_root(dev_queue->qdisc, skb, cb, &q_idx, s_q_idx) < 0) | |
1122 | goto done; | |
1123 | ||
7562f876 PE |
1124 | cont: |
1125 | idx++; | |
1da177e4 LT |
1126 | } |
1127 | ||
1128 | done: | |
1129 | read_unlock(&dev_base_lock); | |
1130 | ||
1131 | cb->args[0] = idx; | |
1132 | cb->args[1] = q_idx; | |
1133 | ||
1134 | return skb->len; | |
1135 | } | |
1136 | ||
1137 | ||
1138 | ||
1139 | /************************************************ | |
1140 | * Traffic classes manipulation. * | |
1141 | ************************************************/ | |
1142 | ||
1143 | ||
1144 | ||
1145 | static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
1146 | { | |
3b1e0a65 | 1147 | struct net *net = sock_net(skb->sk); |
b0e1e646 | 1148 | struct netdev_queue *dev_queue; |
1da177e4 | 1149 | struct tcmsg *tcm = NLMSG_DATA(n); |
1e90474c | 1150 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
1151 | struct net_device *dev; |
1152 | struct Qdisc *q = NULL; | |
20fea08b | 1153 | const struct Qdisc_class_ops *cops; |
1da177e4 LT |
1154 | unsigned long cl = 0; |
1155 | unsigned long new_cl; | |
1156 | u32 pid = tcm->tcm_parent; | |
1157 | u32 clid = tcm->tcm_handle; | |
1158 | u32 qid = TC_H_MAJ(clid); | |
1159 | int err; | |
1160 | ||
b854272b DL |
1161 | if (net != &init_net) |
1162 | return -EINVAL; | |
1163 | ||
881d966b | 1164 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
1165 | return -ENODEV; |
1166 | ||
1e90474c PM |
1167 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); |
1168 | if (err < 0) | |
1169 | return err; | |
1170 | ||
1da177e4 LT |
1171 | /* |
1172 | parent == TC_H_UNSPEC - unspecified parent. | |
1173 | parent == TC_H_ROOT - class is root, which has no parent. | |
1174 | parent == X:0 - parent is root class. | |
1175 | parent == X:Y - parent is a node in hierarchy. | |
1176 | parent == 0:Y - parent is X:Y, where X:0 is qdisc. | |
1177 | ||
1178 | handle == 0:0 - generate handle from kernel pool. | |
1179 | handle == 0:Y - class is X:Y, where X:0 is qdisc. | |
1180 | handle == X:Y - clear. | |
1181 | handle == X:0 - root class. | |
1182 | */ | |
1183 | ||
1184 | /* Step 1. Determine qdisc handle X:0 */ | |
1185 | ||
e8a0464c | 1186 | dev_queue = netdev_get_tx_queue(dev, 0); |
1da177e4 LT |
1187 | if (pid != TC_H_ROOT) { |
1188 | u32 qid1 = TC_H_MAJ(pid); | |
1189 | ||
1190 | if (qid && qid1) { | |
1191 | /* If both majors are known, they must be identical. */ | |
1192 | if (qid != qid1) | |
1193 | return -EINVAL; | |
1194 | } else if (qid1) { | |
1195 | qid = qid1; | |
1196 | } else if (qid == 0) | |
b0e1e646 | 1197 | qid = dev_queue->qdisc_sleeping->handle; |
1da177e4 LT |
1198 | |
1199 | /* Now qid is genuine qdisc handle consistent | |
1200 | both with parent and child. | |
1201 | ||
1202 | TC_H_MAJ(pid) still may be unspecified, complete it now. | |
1203 | */ | |
1204 | if (pid) | |
1205 | pid = TC_H_MAKE(qid, pid); | |
1206 | } else { | |
1207 | if (qid == 0) | |
b0e1e646 | 1208 | qid = dev_queue->qdisc_sleeping->handle; |
1da177e4 LT |
1209 | } |
1210 | ||
1211 | /* OK. Locate qdisc */ | |
10297b99 | 1212 | if ((q = qdisc_lookup(dev, qid)) == NULL) |
1da177e4 LT |
1213 | return -ENOENT; |
1214 | ||
1215 | /* An check that it supports classes */ | |
1216 | cops = q->ops->cl_ops; | |
1217 | if (cops == NULL) | |
1218 | return -EINVAL; | |
1219 | ||
1220 | /* Now try to get class */ | |
1221 | if (clid == 0) { | |
1222 | if (pid == TC_H_ROOT) | |
1223 | clid = qid; | |
1224 | } else | |
1225 | clid = TC_H_MAKE(qid, clid); | |
1226 | ||
1227 | if (clid) | |
1228 | cl = cops->get(q, clid); | |
1229 | ||
1230 | if (cl == 0) { | |
1231 | err = -ENOENT; | |
1232 | if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE)) | |
1233 | goto out; | |
1234 | } else { | |
1235 | switch (n->nlmsg_type) { | |
10297b99 | 1236 | case RTM_NEWTCLASS: |
1da177e4 LT |
1237 | err = -EEXIST; |
1238 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1239 | goto out; | |
1240 | break; | |
1241 | case RTM_DELTCLASS: | |
1242 | err = cops->delete(q, cl); | |
1243 | if (err == 0) | |
1244 | tclass_notify(skb, n, q, cl, RTM_DELTCLASS); | |
1245 | goto out; | |
1246 | case RTM_GETTCLASS: | |
1247 | err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS); | |
1248 | goto out; | |
1249 | default: | |
1250 | err = -EINVAL; | |
1251 | goto out; | |
1252 | } | |
1253 | } | |
1254 | ||
1255 | new_cl = cl; | |
1256 | err = cops->change(q, clid, pid, tca, &new_cl); | |
1257 | if (err == 0) | |
1258 | tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS); | |
1259 | ||
1260 | out: | |
1261 | if (cl) | |
1262 | cops->put(q, cl); | |
1263 | ||
1264 | return err; | |
1265 | } | |
1266 | ||
1267 | ||
1268 | static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, | |
1269 | unsigned long cl, | |
e431b8c0 | 1270 | u32 pid, u32 seq, u16 flags, int event) |
1da177e4 LT |
1271 | { |
1272 | struct tcmsg *tcm; | |
1273 | struct nlmsghdr *nlh; | |
27a884dc | 1274 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 | 1275 | struct gnet_dump d; |
20fea08b | 1276 | const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; |
1da177e4 | 1277 | |
e431b8c0 | 1278 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); |
1da177e4 LT |
1279 | tcm = NLMSG_DATA(nlh); |
1280 | tcm->tcm_family = AF_UNSPEC; | |
5ce2d488 | 1281 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex; |
1da177e4 LT |
1282 | tcm->tcm_parent = q->handle; |
1283 | tcm->tcm_handle = q->handle; | |
1284 | tcm->tcm_info = 0; | |
57e1c487 | 1285 | NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); |
1da177e4 | 1286 | if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) |
1e90474c | 1287 | goto nla_put_failure; |
1da177e4 LT |
1288 | |
1289 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
7698b4fc | 1290 | TCA_XSTATS, qdisc_root_lock(q), &d) < 0) |
1e90474c | 1291 | goto nla_put_failure; |
1da177e4 LT |
1292 | |
1293 | if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) | |
1e90474c | 1294 | goto nla_put_failure; |
1da177e4 LT |
1295 | |
1296 | if (gnet_stats_finish_copy(&d) < 0) | |
1e90474c | 1297 | goto nla_put_failure; |
1da177e4 | 1298 | |
27a884dc | 1299 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
1300 | return skb->len; |
1301 | ||
1302 | nlmsg_failure: | |
1e90474c | 1303 | nla_put_failure: |
dc5fc579 | 1304 | nlmsg_trim(skb, b); |
1da177e4 LT |
1305 | return -1; |
1306 | } | |
1307 | ||
1308 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1309 | struct Qdisc *q, unsigned long cl, int event) | |
1310 | { | |
1311 | struct sk_buff *skb; | |
1312 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1313 | ||
1314 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1315 | if (!skb) | |
1316 | return -ENOBUFS; | |
1317 | ||
1318 | if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) { | |
1319 | kfree_skb(skb); | |
1320 | return -EINVAL; | |
1321 | } | |
1322 | ||
97c53cac | 1323 | return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); |
1da177e4 LT |
1324 | } |
1325 | ||
1326 | struct qdisc_dump_args | |
1327 | { | |
1328 | struct qdisc_walker w; | |
1329 | struct sk_buff *skb; | |
1330 | struct netlink_callback *cb; | |
1331 | }; | |
1332 | ||
1333 | static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) | |
1334 | { | |
1335 | struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; | |
1336 | ||
1337 | return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid, | |
1338 | a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); | |
1339 | } | |
1340 | ||
30723673 DM |
1341 | static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, |
1342 | struct tcmsg *tcm, struct netlink_callback *cb, | |
1343 | int *t_p, int s_t) | |
1344 | { | |
1345 | struct qdisc_dump_args arg; | |
1346 | ||
1347 | if (tc_qdisc_dump_ignore(q) || | |
1348 | *t_p < s_t || !q->ops->cl_ops || | |
1349 | (tcm->tcm_parent && | |
1350 | TC_H_MAJ(tcm->tcm_parent) != q->handle)) { | |
1351 | (*t_p)++; | |
1352 | return 0; | |
1353 | } | |
1354 | if (*t_p > s_t) | |
1355 | memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); | |
1356 | arg.w.fn = qdisc_class_dump; | |
1357 | arg.skb = skb; | |
1358 | arg.cb = cb; | |
1359 | arg.w.stop = 0; | |
1360 | arg.w.skip = cb->args[1]; | |
1361 | arg.w.count = 0; | |
1362 | q->ops->cl_ops->walk(q, &arg.w); | |
1363 | cb->args[1] = arg.w.count; | |
1364 | if (arg.w.stop) | |
1365 | return -1; | |
1366 | (*t_p)++; | |
1367 | return 0; | |
1368 | } | |
1369 | ||
1370 | static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, | |
1371 | struct tcmsg *tcm, struct netlink_callback *cb, | |
1372 | int *t_p, int s_t) | |
1373 | { | |
1374 | struct Qdisc *q; | |
1375 | ||
1376 | if (!root) | |
1377 | return 0; | |
1378 | ||
1379 | if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) | |
1380 | return -1; | |
1381 | ||
1382 | list_for_each_entry(q, &root->list, list) { | |
1383 | if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) | |
1384 | return -1; | |
1385 | } | |
1386 | ||
1387 | return 0; | |
1388 | } | |
1389 | ||
1da177e4 LT |
1390 | static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) |
1391 | { | |
30723673 | 1392 | struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); |
3b1e0a65 | 1393 | struct net *net = sock_net(skb->sk); |
30723673 | 1394 | struct netdev_queue *dev_queue; |
1da177e4 | 1395 | struct net_device *dev; |
30723673 | 1396 | int t, s_t; |
1da177e4 | 1397 | |
b854272b DL |
1398 | if (net != &init_net) |
1399 | return 0; | |
1400 | ||
1da177e4 LT |
1401 | if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) |
1402 | return 0; | |
881d966b | 1403 | if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) |
1da177e4 LT |
1404 | return 0; |
1405 | ||
1406 | s_t = cb->args[0]; | |
1407 | t = 0; | |
1408 | ||
30723673 DM |
1409 | dev_queue = netdev_get_tx_queue(dev, 0); |
1410 | if (tc_dump_tclass_root(dev_queue->qdisc, skb, tcm, cb, &t, s_t) < 0) | |
1411 | goto done; | |
1412 | ||
1413 | dev_queue = &dev->rx_queue; | |
1414 | if (tc_dump_tclass_root(dev_queue->qdisc, skb, tcm, cb, &t, s_t) < 0) | |
1415 | goto done; | |
1da177e4 | 1416 | |
30723673 | 1417 | done: |
1da177e4 LT |
1418 | cb->args[0] = t; |
1419 | ||
1420 | dev_put(dev); | |
1421 | return skb->len; | |
1422 | } | |
1423 | ||
1424 | /* Main classifier routine: scans classifier chain attached | |
1425 | to this qdisc, (optionally) tests for protocol and asks | |
1426 | specific classifiers. | |
1427 | */ | |
73ca4918 PM |
1428 | int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp, |
1429 | struct tcf_result *res) | |
1430 | { | |
1431 | __be16 protocol = skb->protocol; | |
1432 | int err = 0; | |
1433 | ||
1434 | for (; tp; tp = tp->next) { | |
1435 | if ((tp->protocol == protocol || | |
1436 | tp->protocol == htons(ETH_P_ALL)) && | |
1437 | (err = tp->classify(skb, tp, res)) >= 0) { | |
1438 | #ifdef CONFIG_NET_CLS_ACT | |
1439 | if (err != TC_ACT_RECLASSIFY && skb->tc_verd) | |
1440 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0); | |
1441 | #endif | |
1442 | return err; | |
1443 | } | |
1444 | } | |
1445 | return -1; | |
1446 | } | |
1447 | EXPORT_SYMBOL(tc_classify_compat); | |
1448 | ||
1da177e4 | 1449 | int tc_classify(struct sk_buff *skb, struct tcf_proto *tp, |
73ca4918 | 1450 | struct tcf_result *res) |
1da177e4 LT |
1451 | { |
1452 | int err = 0; | |
73ca4918 | 1453 | __be16 protocol; |
1da177e4 LT |
1454 | #ifdef CONFIG_NET_CLS_ACT |
1455 | struct tcf_proto *otp = tp; | |
1456 | reclassify: | |
1457 | #endif | |
1458 | protocol = skb->protocol; | |
1459 | ||
73ca4918 | 1460 | err = tc_classify_compat(skb, tp, res); |
1da177e4 | 1461 | #ifdef CONFIG_NET_CLS_ACT |
73ca4918 PM |
1462 | if (err == TC_ACT_RECLASSIFY) { |
1463 | u32 verd = G_TC_VERD(skb->tc_verd); | |
1464 | tp = otp; | |
1465 | ||
1466 | if (verd++ >= MAX_REC_LOOP) { | |
1467 | printk("rule prio %u protocol %02x reclassify loop, " | |
1468 | "packet dropped\n", | |
1469 | tp->prio&0xffff, ntohs(tp->protocol)); | |
1470 | return TC_ACT_SHOT; | |
1da177e4 | 1471 | } |
73ca4918 PM |
1472 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd); |
1473 | goto reclassify; | |
1da177e4 | 1474 | } |
73ca4918 PM |
1475 | #endif |
1476 | return err; | |
1da177e4 | 1477 | } |
73ca4918 | 1478 | EXPORT_SYMBOL(tc_classify); |
1da177e4 | 1479 | |
a48b5a61 PM |
1480 | void tcf_destroy(struct tcf_proto *tp) |
1481 | { | |
1482 | tp->ops->destroy(tp); | |
1483 | module_put(tp->ops->owner); | |
1484 | kfree(tp); | |
1485 | } | |
1486 | ||
ff31ab56 | 1487 | void tcf_destroy_chain(struct tcf_proto **fl) |
a48b5a61 PM |
1488 | { |
1489 | struct tcf_proto *tp; | |
1490 | ||
ff31ab56 PM |
1491 | while ((tp = *fl) != NULL) { |
1492 | *fl = tp->next; | |
a48b5a61 PM |
1493 | tcf_destroy(tp); |
1494 | } | |
1495 | } | |
1496 | EXPORT_SYMBOL(tcf_destroy_chain); | |
1497 | ||
1da177e4 LT |
1498 | #ifdef CONFIG_PROC_FS |
1499 | static int psched_show(struct seq_file *seq, void *v) | |
1500 | { | |
3c0cfc13 PM |
1501 | struct timespec ts; |
1502 | ||
1503 | hrtimer_get_res(CLOCK_MONOTONIC, &ts); | |
1da177e4 | 1504 | seq_printf(seq, "%08x %08x %08x %08x\n", |
641b9e0e | 1505 | (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1), |
514bca32 | 1506 | 1000000, |
3c0cfc13 | 1507 | (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts))); |
1da177e4 LT |
1508 | |
1509 | return 0; | |
1510 | } | |
1511 | ||
1512 | static int psched_open(struct inode *inode, struct file *file) | |
1513 | { | |
1514 | return single_open(file, psched_show, PDE(inode)->data); | |
1515 | } | |
1516 | ||
da7071d7 | 1517 | static const struct file_operations psched_fops = { |
1da177e4 LT |
1518 | .owner = THIS_MODULE, |
1519 | .open = psched_open, | |
1520 | .read = seq_read, | |
1521 | .llseek = seq_lseek, | |
1522 | .release = single_release, | |
10297b99 | 1523 | }; |
1da177e4 LT |
1524 | #endif |
1525 | ||
1da177e4 LT |
1526 | static int __init pktsched_init(void) |
1527 | { | |
1da177e4 LT |
1528 | register_qdisc(&pfifo_qdisc_ops); |
1529 | register_qdisc(&bfifo_qdisc_ops); | |
457c4cbc | 1530 | proc_net_fops_create(&init_net, "psched", 0, &psched_fops); |
1da177e4 | 1531 | |
be577ddc TG |
1532 | rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL); |
1533 | rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL); | |
1534 | rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc); | |
1535 | rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL); | |
1536 | rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL); | |
1537 | rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass); | |
1538 | ||
1da177e4 LT |
1539 | return 0; |
1540 | } | |
1541 | ||
1542 | subsys_initcall(pktsched_init); |