]>
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> |
25bfcd5a | 30 | #include <linux/lockdep.h> |
5a0e3ad6 | 31 | #include <linux/slab.h> |
59cc1f61 | 32 | #include <linux/hashtable.h> |
1da177e4 | 33 | |
457c4cbc | 34 | #include <net/net_namespace.h> |
b854272b | 35 | #include <net/sock.h> |
dc5fc579 | 36 | #include <net/netlink.h> |
1da177e4 LT |
37 | #include <net/pkt_sched.h> |
38 | ||
7316ae88 TG |
39 | static int qdisc_notify(struct net *net, struct sk_buff *oskb, |
40 | struct nlmsghdr *n, u32 clid, | |
1da177e4 | 41 | struct Qdisc *old, struct Qdisc *new); |
7316ae88 TG |
42 | static int tclass_notify(struct net *net, struct sk_buff *oskb, |
43 | struct nlmsghdr *n, struct Qdisc *q, | |
44 | unsigned long cl, int event); | |
1da177e4 LT |
45 | |
46 | /* | |
47 | ||
48 | Short review. | |
49 | ------------- | |
50 | ||
51 | This file consists of two interrelated parts: | |
52 | ||
53 | 1. queueing disciplines manager frontend. | |
54 | 2. traffic classes manager frontend. | |
55 | ||
56 | Generally, queueing discipline ("qdisc") is a black box, | |
57 | which is able to enqueue packets and to dequeue them (when | |
58 | device is ready to send something) in order and at times | |
59 | determined by algorithm hidden in it. | |
60 | ||
61 | qdisc's are divided to two categories: | |
62 | - "queues", which have no internal structure visible from outside. | |
63 | - "schedulers", which split all the packets to "traffic classes", | |
64 | using "packet classifiers" (look at cls_api.c) | |
65 | ||
66 | In turn, classes may have child qdiscs (as rule, queues) | |
67 | attached to them etc. etc. etc. | |
68 | ||
69 | The goal of the routines in this file is to translate | |
70 | information supplied by user in the form of handles | |
71 | to more intelligible for kernel form, to make some sanity | |
72 | checks and part of work, which is common to all qdiscs | |
73 | and to provide rtnetlink notifications. | |
74 | ||
75 | All real intelligent work is done inside qdisc modules. | |
76 | ||
77 | ||
78 | ||
79 | Every discipline has two major routines: enqueue and dequeue. | |
80 | ||
81 | ---dequeue | |
82 | ||
83 | dequeue usually returns a skb to send. It is allowed to return NULL, | |
84 | but it does not mean that queue is empty, it just means that | |
85 | discipline does not want to send anything this time. | |
86 | Queue is really empty if q->q.qlen == 0. | |
87 | For complicated disciplines with multiple queues q->q is not | |
88 | real packet queue, but however q->q.qlen must be valid. | |
89 | ||
90 | ---enqueue | |
91 | ||
92 | enqueue returns 0, if packet was enqueued successfully. | |
93 | If packet (this one or another one) was dropped, it returns | |
94 | not zero error code. | |
95 | NET_XMIT_DROP - this packet dropped | |
96 | Expected action: do not backoff, but wait until queue will clear. | |
97 | NET_XMIT_CN - probably this packet enqueued, but another one dropped. | |
98 | Expected action: backoff or ignore | |
1da177e4 LT |
99 | |
100 | Auxiliary routines: | |
101 | ||
99c0db26 JP |
102 | ---peek |
103 | ||
104 | like dequeue but without removing a packet from the queue | |
105 | ||
1da177e4 LT |
106 | ---reset |
107 | ||
108 | returns qdisc to initial state: purge all buffers, clear all | |
109 | timers, counters (except for statistics) etc. | |
110 | ||
111 | ---init | |
112 | ||
113 | initializes newly created qdisc. | |
114 | ||
115 | ---destroy | |
116 | ||
117 | destroys resources allocated by init and during lifetime of qdisc. | |
118 | ||
119 | ---change | |
120 | ||
121 | changes qdisc parameters. | |
122 | */ | |
123 | ||
124 | /* Protects list of registered TC modules. It is pure SMP lock. */ | |
125 | static DEFINE_RWLOCK(qdisc_mod_lock); | |
126 | ||
127 | ||
128 | /************************************************ | |
129 | * Queueing disciplines manipulation. * | |
130 | ************************************************/ | |
131 | ||
132 | ||
133 | /* The list of all installed queueing disciplines. */ | |
134 | ||
135 | static struct Qdisc_ops *qdisc_base; | |
136 | ||
21eb2189 | 137 | /* Register/unregister queueing discipline */ |
1da177e4 LT |
138 | |
139 | int register_qdisc(struct Qdisc_ops *qops) | |
140 | { | |
141 | struct Qdisc_ops *q, **qp; | |
142 | int rc = -EEXIST; | |
143 | ||
144 | write_lock(&qdisc_mod_lock); | |
145 | for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) | |
146 | if (!strcmp(qops->id, q->id)) | |
147 | goto out; | |
148 | ||
149 | if (qops->enqueue == NULL) | |
150 | qops->enqueue = noop_qdisc_ops.enqueue; | |
99c0db26 | 151 | if (qops->peek == NULL) { |
68fd26b5 | 152 | if (qops->dequeue == NULL) |
99c0db26 | 153 | qops->peek = noop_qdisc_ops.peek; |
68fd26b5 JP |
154 | else |
155 | goto out_einval; | |
99c0db26 | 156 | } |
1da177e4 LT |
157 | if (qops->dequeue == NULL) |
158 | qops->dequeue = noop_qdisc_ops.dequeue; | |
159 | ||
68fd26b5 JP |
160 | if (qops->cl_ops) { |
161 | const struct Qdisc_class_ops *cops = qops->cl_ops; | |
162 | ||
3e9e5a59 | 163 | if (!(cops->get && cops->put && cops->walk && cops->leaf)) |
68fd26b5 JP |
164 | goto out_einval; |
165 | ||
166 | if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf)) | |
167 | goto out_einval; | |
168 | } | |
169 | ||
1da177e4 LT |
170 | qops->next = NULL; |
171 | *qp = qops; | |
172 | rc = 0; | |
173 | out: | |
174 | write_unlock(&qdisc_mod_lock); | |
175 | return rc; | |
68fd26b5 JP |
176 | |
177 | out_einval: | |
178 | rc = -EINVAL; | |
179 | goto out; | |
1da177e4 | 180 | } |
62e3ba1b | 181 | EXPORT_SYMBOL(register_qdisc); |
1da177e4 LT |
182 | |
183 | int unregister_qdisc(struct Qdisc_ops *qops) | |
184 | { | |
185 | struct Qdisc_ops *q, **qp; | |
186 | int err = -ENOENT; | |
187 | ||
188 | write_lock(&qdisc_mod_lock); | |
cc7ec456 | 189 | for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) |
1da177e4 LT |
190 | if (q == qops) |
191 | break; | |
192 | if (q) { | |
193 | *qp = q->next; | |
194 | q->next = NULL; | |
195 | err = 0; | |
196 | } | |
197 | write_unlock(&qdisc_mod_lock); | |
198 | return err; | |
199 | } | |
62e3ba1b | 200 | EXPORT_SYMBOL(unregister_qdisc); |
1da177e4 | 201 | |
6da7c8fc | 202 | /* Get default qdisc if not otherwise specified */ |
203 | void qdisc_get_default(char *name, size_t len) | |
204 | { | |
205 | read_lock(&qdisc_mod_lock); | |
206 | strlcpy(name, default_qdisc_ops->id, len); | |
207 | read_unlock(&qdisc_mod_lock); | |
208 | } | |
209 | ||
210 | static struct Qdisc_ops *qdisc_lookup_default(const char *name) | |
211 | { | |
212 | struct Qdisc_ops *q = NULL; | |
213 | ||
214 | for (q = qdisc_base; q; q = q->next) { | |
215 | if (!strcmp(name, q->id)) { | |
216 | if (!try_module_get(q->owner)) | |
217 | q = NULL; | |
218 | break; | |
219 | } | |
220 | } | |
221 | ||
222 | return q; | |
223 | } | |
224 | ||
225 | /* Set new default qdisc to use */ | |
226 | int qdisc_set_default(const char *name) | |
227 | { | |
228 | const struct Qdisc_ops *ops; | |
229 | ||
230 | if (!capable(CAP_NET_ADMIN)) | |
231 | return -EPERM; | |
232 | ||
233 | write_lock(&qdisc_mod_lock); | |
234 | ops = qdisc_lookup_default(name); | |
235 | if (!ops) { | |
236 | /* Not found, drop lock and try to load module */ | |
237 | write_unlock(&qdisc_mod_lock); | |
238 | request_module("sch_%s", name); | |
239 | write_lock(&qdisc_mod_lock); | |
240 | ||
241 | ops = qdisc_lookup_default(name); | |
242 | } | |
243 | ||
244 | if (ops) { | |
245 | /* Set new default */ | |
246 | module_put(default_qdisc_ops->owner); | |
247 | default_qdisc_ops = ops; | |
248 | } | |
249 | write_unlock(&qdisc_mod_lock); | |
250 | ||
251 | return ops ? 0 : -ENOENT; | |
252 | } | |
253 | ||
8ea3e439 | 254 | #ifdef CONFIG_NET_SCH_DEFAULT |
255 | /* Set default value from kernel config */ | |
256 | static int __init sch_default_qdisc(void) | |
257 | { | |
258 | return qdisc_set_default(CONFIG_DEFAULT_NET_SCH); | |
259 | } | |
260 | late_initcall(sch_default_qdisc); | |
261 | #endif | |
262 | ||
1da177e4 | 263 | /* We know handle. Find qdisc among all qdisc's attached to device |
4eaf3b84 ED |
264 | * (root qdisc, all its children, children of children etc.) |
265 | * Note: caller either uses rtnl or rcu_read_lock() | |
1da177e4 LT |
266 | */ |
267 | ||
6113b748 | 268 | static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle) |
8123b421 DM |
269 | { |
270 | struct Qdisc *q; | |
271 | ||
69012ae4 JK |
272 | if (!qdisc_dev(root)) |
273 | return (root->handle == handle ? root : NULL); | |
274 | ||
8123b421 DM |
275 | if (!(root->flags & TCQ_F_BUILTIN) && |
276 | root->handle == handle) | |
277 | return root; | |
278 | ||
59cc1f61 | 279 | hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) { |
8123b421 DM |
280 | if (q->handle == handle) |
281 | return q; | |
282 | } | |
283 | return NULL; | |
284 | } | |
285 | ||
49b49971 | 286 | void qdisc_hash_add(struct Qdisc *q, bool invisible) |
f6e0b239 | 287 | { |
37314363 ED |
288 | if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { |
289 | struct Qdisc *root = qdisc_dev(q)->qdisc; | |
e57a784d | 290 | |
37314363 | 291 | WARN_ON_ONCE(root == &noop_qdisc); |
4eaf3b84 | 292 | ASSERT_RTNL(); |
59cc1f61 | 293 | hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle); |
49b49971 JK |
294 | if (invisible) |
295 | q->flags |= TCQ_F_INVISIBLE; | |
37314363 | 296 | } |
f6e0b239 | 297 | } |
59cc1f61 | 298 | EXPORT_SYMBOL(qdisc_hash_add); |
f6e0b239 | 299 | |
59cc1f61 | 300 | void qdisc_hash_del(struct Qdisc *q) |
f6e0b239 | 301 | { |
4eaf3b84 ED |
302 | if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { |
303 | ASSERT_RTNL(); | |
59cc1f61 | 304 | hash_del_rcu(&q->hash); |
4eaf3b84 | 305 | } |
f6e0b239 | 306 | } |
59cc1f61 | 307 | EXPORT_SYMBOL(qdisc_hash_del); |
f6e0b239 | 308 | |
ead81cc5 | 309 | struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) |
1da177e4 | 310 | { |
f6e0b239 JP |
311 | struct Qdisc *q; |
312 | ||
af356afa PM |
313 | q = qdisc_match_from_root(dev->qdisc, handle); |
314 | if (q) | |
315 | goto out; | |
f6e0b239 | 316 | |
24824a09 ED |
317 | if (dev_ingress_queue(dev)) |
318 | q = qdisc_match_from_root( | |
319 | dev_ingress_queue(dev)->qdisc_sleeping, | |
320 | handle); | |
f6486d40 | 321 | out: |
f6e0b239 | 322 | return q; |
1da177e4 LT |
323 | } |
324 | ||
325 | static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) | |
326 | { | |
327 | unsigned long cl; | |
328 | struct Qdisc *leaf; | |
20fea08b | 329 | const struct Qdisc_class_ops *cops = p->ops->cl_ops; |
1da177e4 LT |
330 | |
331 | if (cops == NULL) | |
332 | return NULL; | |
333 | cl = cops->get(p, classid); | |
334 | ||
335 | if (cl == 0) | |
336 | return NULL; | |
337 | leaf = cops->leaf(p, cl); | |
338 | cops->put(p, cl); | |
339 | return leaf; | |
340 | } | |
341 | ||
342 | /* Find queueing discipline by name */ | |
343 | ||
1e90474c | 344 | static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) |
1da177e4 LT |
345 | { |
346 | struct Qdisc_ops *q = NULL; | |
347 | ||
348 | if (kind) { | |
349 | read_lock(&qdisc_mod_lock); | |
350 | for (q = qdisc_base; q; q = q->next) { | |
1e90474c | 351 | if (nla_strcmp(kind, q->id) == 0) { |
1da177e4 LT |
352 | if (!try_module_get(q->owner)) |
353 | q = NULL; | |
354 | break; | |
355 | } | |
356 | } | |
357 | read_unlock(&qdisc_mod_lock); | |
358 | } | |
359 | return q; | |
360 | } | |
361 | ||
8a8e3d84 JDB |
362 | /* The linklayer setting were not transferred from iproute2, in older |
363 | * versions, and the rate tables lookup systems have been dropped in | |
364 | * the kernel. To keep backward compatible with older iproute2 tc | |
365 | * utils, we detect the linklayer setting by detecting if the rate | |
366 | * table were modified. | |
367 | * | |
368 | * For linklayer ATM table entries, the rate table will be aligned to | |
369 | * 48 bytes, thus some table entries will contain the same value. The | |
370 | * mpu (min packet unit) is also encoded into the old rate table, thus | |
371 | * starting from the mpu, we find low and high table entries for | |
372 | * mapping this cell. If these entries contain the same value, when | |
373 | * the rate tables have been modified for linklayer ATM. | |
374 | * | |
375 | * This is done by rounding mpu to the nearest 48 bytes cell/entry, | |
376 | * and then roundup to the next cell, calc the table entry one below, | |
377 | * and compare. | |
378 | */ | |
379 | static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab) | |
380 | { | |
381 | int low = roundup(r->mpu, 48); | |
382 | int high = roundup(low+1, 48); | |
383 | int cell_low = low >> r->cell_log; | |
384 | int cell_high = (high >> r->cell_log) - 1; | |
385 | ||
386 | /* rtab is too inaccurate at rates > 100Mbit/s */ | |
387 | if ((r->rate > (100000000/8)) || (rtab[0] == 0)) { | |
388 | pr_debug("TC linklayer: Giving up ATM detection\n"); | |
389 | return TC_LINKLAYER_ETHERNET; | |
390 | } | |
391 | ||
392 | if ((cell_high > cell_low) && (cell_high < 256) | |
393 | && (rtab[cell_low] == rtab[cell_high])) { | |
394 | pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n", | |
395 | cell_low, cell_high, rtab[cell_high]); | |
396 | return TC_LINKLAYER_ATM; | |
397 | } | |
398 | return TC_LINKLAYER_ETHERNET; | |
399 | } | |
400 | ||
1da177e4 LT |
401 | static struct qdisc_rate_table *qdisc_rtab_list; |
402 | ||
5a7a5555 JHS |
403 | struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, |
404 | struct nlattr *tab) | |
1da177e4 LT |
405 | { |
406 | struct qdisc_rate_table *rtab; | |
407 | ||
40edeff6 ED |
408 | if (tab == NULL || r->rate == 0 || r->cell_log == 0 || |
409 | nla_len(tab) != TC_RTAB_SIZE) | |
410 | return NULL; | |
411 | ||
1da177e4 | 412 | for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { |
40edeff6 ED |
413 | if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) && |
414 | !memcmp(&rtab->data, nla_data(tab), 1024)) { | |
1da177e4 LT |
415 | rtab->refcnt++; |
416 | return rtab; | |
417 | } | |
418 | } | |
419 | ||
1da177e4 LT |
420 | rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); |
421 | if (rtab) { | |
422 | rtab->rate = *r; | |
423 | rtab->refcnt = 1; | |
1e90474c | 424 | memcpy(rtab->data, nla_data(tab), 1024); |
8a8e3d84 JDB |
425 | if (r->linklayer == TC_LINKLAYER_UNAWARE) |
426 | r->linklayer = __detect_linklayer(r, rtab->data); | |
1da177e4 LT |
427 | rtab->next = qdisc_rtab_list; |
428 | qdisc_rtab_list = rtab; | |
429 | } | |
430 | return rtab; | |
431 | } | |
62e3ba1b | 432 | EXPORT_SYMBOL(qdisc_get_rtab); |
1da177e4 LT |
433 | |
434 | void qdisc_put_rtab(struct qdisc_rate_table *tab) | |
435 | { | |
436 | struct qdisc_rate_table *rtab, **rtabp; | |
437 | ||
438 | if (!tab || --tab->refcnt) | |
439 | return; | |
440 | ||
cc7ec456 ED |
441 | for (rtabp = &qdisc_rtab_list; |
442 | (rtab = *rtabp) != NULL; | |
443 | rtabp = &rtab->next) { | |
1da177e4 LT |
444 | if (rtab == tab) { |
445 | *rtabp = rtab->next; | |
446 | kfree(rtab); | |
447 | return; | |
448 | } | |
449 | } | |
450 | } | |
62e3ba1b | 451 | EXPORT_SYMBOL(qdisc_put_rtab); |
1da177e4 | 452 | |
175f9c1b | 453 | static LIST_HEAD(qdisc_stab_list); |
175f9c1b JK |
454 | |
455 | static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { | |
456 | [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, | |
457 | [TCA_STAB_DATA] = { .type = NLA_BINARY }, | |
458 | }; | |
459 | ||
460 | static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) | |
461 | { | |
462 | struct nlattr *tb[TCA_STAB_MAX + 1]; | |
463 | struct qdisc_size_table *stab; | |
464 | struct tc_sizespec *s; | |
465 | unsigned int tsize = 0; | |
466 | u16 *tab = NULL; | |
467 | int err; | |
468 | ||
fceb6435 | 469 | err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy, NULL); |
175f9c1b JK |
470 | if (err < 0) |
471 | return ERR_PTR(err); | |
472 | if (!tb[TCA_STAB_BASE]) | |
473 | return ERR_PTR(-EINVAL); | |
474 | ||
475 | s = nla_data(tb[TCA_STAB_BASE]); | |
476 | ||
477 | if (s->tsize > 0) { | |
478 | if (!tb[TCA_STAB_DATA]) | |
479 | return ERR_PTR(-EINVAL); | |
480 | tab = nla_data(tb[TCA_STAB_DATA]); | |
481 | tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); | |
482 | } | |
483 | ||
00093fab | 484 | if (tsize != s->tsize || (!tab && tsize > 0)) |
175f9c1b JK |
485 | return ERR_PTR(-EINVAL); |
486 | ||
175f9c1b JK |
487 | list_for_each_entry(stab, &qdisc_stab_list, list) { |
488 | if (memcmp(&stab->szopts, s, sizeof(*s))) | |
489 | continue; | |
490 | if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16))) | |
491 | continue; | |
492 | stab->refcnt++; | |
175f9c1b JK |
493 | return stab; |
494 | } | |
495 | ||
175f9c1b JK |
496 | stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); |
497 | if (!stab) | |
498 | return ERR_PTR(-ENOMEM); | |
499 | ||
500 | stab->refcnt = 1; | |
501 | stab->szopts = *s; | |
502 | if (tsize > 0) | |
503 | memcpy(stab->data, tab, tsize * sizeof(u16)); | |
504 | ||
175f9c1b | 505 | list_add_tail(&stab->list, &qdisc_stab_list); |
175f9c1b JK |
506 | |
507 | return stab; | |
508 | } | |
509 | ||
a2da570d ED |
510 | static void stab_kfree_rcu(struct rcu_head *head) |
511 | { | |
512 | kfree(container_of(head, struct qdisc_size_table, rcu)); | |
513 | } | |
514 | ||
175f9c1b JK |
515 | void qdisc_put_stab(struct qdisc_size_table *tab) |
516 | { | |
517 | if (!tab) | |
518 | return; | |
519 | ||
175f9c1b JK |
520 | if (--tab->refcnt == 0) { |
521 | list_del(&tab->list); | |
a2da570d | 522 | call_rcu_bh(&tab->rcu, stab_kfree_rcu); |
175f9c1b | 523 | } |
175f9c1b JK |
524 | } |
525 | EXPORT_SYMBOL(qdisc_put_stab); | |
526 | ||
527 | static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) | |
528 | { | |
529 | struct nlattr *nest; | |
530 | ||
531 | nest = nla_nest_start(skb, TCA_STAB); | |
3aa4614d PM |
532 | if (nest == NULL) |
533 | goto nla_put_failure; | |
1b34ec43 DM |
534 | if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts)) |
535 | goto nla_put_failure; | |
175f9c1b JK |
536 | nla_nest_end(skb, nest); |
537 | ||
538 | return skb->len; | |
539 | ||
540 | nla_put_failure: | |
541 | return -1; | |
542 | } | |
543 | ||
5a7a5555 JHS |
544 | void __qdisc_calculate_pkt_len(struct sk_buff *skb, |
545 | const struct qdisc_size_table *stab) | |
175f9c1b JK |
546 | { |
547 | int pkt_len, slot; | |
548 | ||
549 | pkt_len = skb->len + stab->szopts.overhead; | |
550 | if (unlikely(!stab->szopts.tsize)) | |
551 | goto out; | |
552 | ||
553 | slot = pkt_len + stab->szopts.cell_align; | |
554 | if (unlikely(slot < 0)) | |
555 | slot = 0; | |
556 | ||
557 | slot >>= stab->szopts.cell_log; | |
558 | if (likely(slot < stab->szopts.tsize)) | |
559 | pkt_len = stab->data[slot]; | |
560 | else | |
561 | pkt_len = stab->data[stab->szopts.tsize - 1] * | |
562 | (slot / stab->szopts.tsize) + | |
563 | stab->data[slot % stab->szopts.tsize]; | |
564 | ||
565 | pkt_len <<= stab->szopts.size_log; | |
566 | out: | |
567 | if (unlikely(pkt_len < 1)) | |
568 | pkt_len = 1; | |
569 | qdisc_skb_cb(skb)->pkt_len = pkt_len; | |
570 | } | |
a2da570d | 571 | EXPORT_SYMBOL(__qdisc_calculate_pkt_len); |
175f9c1b | 572 | |
6e765a00 | 573 | void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc) |
b00355db JP |
574 | { |
575 | if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { | |
cc7ec456 ED |
576 | pr_warn("%s: %s qdisc %X: is non-work-conserving?\n", |
577 | txt, qdisc->ops->id, qdisc->handle >> 16); | |
b00355db JP |
578 | qdisc->flags |= TCQ_F_WARN_NONWC; |
579 | } | |
580 | } | |
581 | EXPORT_SYMBOL(qdisc_warn_nonwc); | |
582 | ||
4179477f PM |
583 | static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) |
584 | { | |
585 | struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, | |
2fbd3da3 | 586 | timer); |
4179477f | 587 | |
1e203c1a | 588 | rcu_read_lock(); |
8608db03 | 589 | __netif_schedule(qdisc_root(wd->qdisc)); |
1e203c1a | 590 | rcu_read_unlock(); |
1936502d | 591 | |
4179477f PM |
592 | return HRTIMER_NORESTART; |
593 | } | |
594 | ||
595 | void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) | |
596 | { | |
4a8e320c | 597 | hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED); |
2fbd3da3 | 598 | wd->timer.function = qdisc_watchdog; |
4179477f PM |
599 | wd->qdisc = qdisc; |
600 | } | |
601 | EXPORT_SYMBOL(qdisc_watchdog_init); | |
602 | ||
45f50bed | 603 | void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires) |
4179477f | 604 | { |
2540e051 JP |
605 | if (test_bit(__QDISC_STATE_DEACTIVATED, |
606 | &qdisc_root_sleeping(wd->qdisc)->state)) | |
607 | return; | |
608 | ||
a9efad8b ED |
609 | if (wd->last_expires == expires) |
610 | return; | |
611 | ||
612 | wd->last_expires = expires; | |
46baac38 | 613 | hrtimer_start(&wd->timer, |
34c5d292 | 614 | ns_to_ktime(expires), |
4a8e320c | 615 | HRTIMER_MODE_ABS_PINNED); |
4179477f | 616 | } |
34c5d292 | 617 | EXPORT_SYMBOL(qdisc_watchdog_schedule_ns); |
4179477f PM |
618 | |
619 | void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) | |
620 | { | |
2fbd3da3 | 621 | hrtimer_cancel(&wd->timer); |
4179477f PM |
622 | } |
623 | EXPORT_SYMBOL(qdisc_watchdog_cancel); | |
1da177e4 | 624 | |
a94f779f | 625 | static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) |
6fe1c7a5 PM |
626 | { |
627 | unsigned int size = n * sizeof(struct hlist_head), i; | |
628 | struct hlist_head *h; | |
629 | ||
630 | if (size <= PAGE_SIZE) | |
631 | h = kmalloc(size, GFP_KERNEL); | |
632 | else | |
633 | h = (struct hlist_head *) | |
634 | __get_free_pages(GFP_KERNEL, get_order(size)); | |
635 | ||
636 | if (h != NULL) { | |
637 | for (i = 0; i < n; i++) | |
638 | INIT_HLIST_HEAD(&h[i]); | |
639 | } | |
640 | return h; | |
641 | } | |
642 | ||
643 | static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) | |
644 | { | |
645 | unsigned int size = n * sizeof(struct hlist_head); | |
646 | ||
647 | if (size <= PAGE_SIZE) | |
648 | kfree(h); | |
649 | else | |
650 | free_pages((unsigned long)h, get_order(size)); | |
651 | } | |
652 | ||
653 | void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) | |
654 | { | |
655 | struct Qdisc_class_common *cl; | |
b67bfe0d | 656 | struct hlist_node *next; |
6fe1c7a5 PM |
657 | struct hlist_head *nhash, *ohash; |
658 | unsigned int nsize, nmask, osize; | |
659 | unsigned int i, h; | |
660 | ||
661 | /* Rehash when load factor exceeds 0.75 */ | |
662 | if (clhash->hashelems * 4 <= clhash->hashsize * 3) | |
663 | return; | |
664 | nsize = clhash->hashsize * 2; | |
665 | nmask = nsize - 1; | |
666 | nhash = qdisc_class_hash_alloc(nsize); | |
667 | if (nhash == NULL) | |
668 | return; | |
669 | ||
670 | ohash = clhash->hash; | |
671 | osize = clhash->hashsize; | |
672 | ||
673 | sch_tree_lock(sch); | |
674 | for (i = 0; i < osize; i++) { | |
b67bfe0d | 675 | hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) { |
6fe1c7a5 PM |
676 | h = qdisc_class_hash(cl->classid, nmask); |
677 | hlist_add_head(&cl->hnode, &nhash[h]); | |
678 | } | |
679 | } | |
680 | clhash->hash = nhash; | |
681 | clhash->hashsize = nsize; | |
682 | clhash->hashmask = nmask; | |
683 | sch_tree_unlock(sch); | |
684 | ||
685 | qdisc_class_hash_free(ohash, osize); | |
686 | } | |
687 | EXPORT_SYMBOL(qdisc_class_hash_grow); | |
688 | ||
689 | int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) | |
690 | { | |
691 | unsigned int size = 4; | |
692 | ||
693 | clhash->hash = qdisc_class_hash_alloc(size); | |
694 | if (clhash->hash == NULL) | |
695 | return -ENOMEM; | |
696 | clhash->hashsize = size; | |
697 | clhash->hashmask = size - 1; | |
698 | clhash->hashelems = 0; | |
699 | return 0; | |
700 | } | |
701 | EXPORT_SYMBOL(qdisc_class_hash_init); | |
702 | ||
703 | void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) | |
704 | { | |
705 | qdisc_class_hash_free(clhash->hash, clhash->hashsize); | |
706 | } | |
707 | EXPORT_SYMBOL(qdisc_class_hash_destroy); | |
708 | ||
709 | void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, | |
710 | struct Qdisc_class_common *cl) | |
711 | { | |
712 | unsigned int h; | |
713 | ||
714 | INIT_HLIST_NODE(&cl->hnode); | |
715 | h = qdisc_class_hash(cl->classid, clhash->hashmask); | |
716 | hlist_add_head(&cl->hnode, &clhash->hash[h]); | |
717 | clhash->hashelems++; | |
718 | } | |
719 | EXPORT_SYMBOL(qdisc_class_hash_insert); | |
720 | ||
721 | void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, | |
722 | struct Qdisc_class_common *cl) | |
723 | { | |
724 | hlist_del(&cl->hnode); | |
725 | clhash->hashelems--; | |
726 | } | |
727 | EXPORT_SYMBOL(qdisc_class_hash_remove); | |
728 | ||
fa0f5aa7 ED |
729 | /* Allocate an unique handle from space managed by kernel |
730 | * Possible range is [8000-FFFF]:0000 (0x8000 values) | |
731 | */ | |
1da177e4 LT |
732 | static u32 qdisc_alloc_handle(struct net_device *dev) |
733 | { | |
fa0f5aa7 | 734 | int i = 0x8000; |
1da177e4 LT |
735 | static u32 autohandle = TC_H_MAKE(0x80000000U, 0); |
736 | ||
737 | do { | |
738 | autohandle += TC_H_MAKE(0x10000U, 0); | |
739 | if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) | |
740 | autohandle = TC_H_MAKE(0x80000000U, 0); | |
fa0f5aa7 ED |
741 | if (!qdisc_lookup(dev, autohandle)) |
742 | return autohandle; | |
743 | cond_resched(); | |
744 | } while (--i > 0); | |
1da177e4 | 745 | |
fa0f5aa7 | 746 | return 0; |
1da177e4 LT |
747 | } |
748 | ||
2ccccf5f WC |
749 | void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n, |
750 | unsigned int len) | |
43effa1e | 751 | { |
20fea08b | 752 | const struct Qdisc_class_ops *cops; |
43effa1e PM |
753 | unsigned long cl; |
754 | u32 parentid; | |
2c8c8e6f | 755 | int drops; |
43effa1e | 756 | |
2ccccf5f | 757 | if (n == 0 && len == 0) |
43effa1e | 758 | return; |
2c8c8e6f | 759 | drops = max_t(int, n, 0); |
4eaf3b84 | 760 | rcu_read_lock(); |
43effa1e | 761 | while ((parentid = sch->parent)) { |
066a3b5b | 762 | if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) |
4eaf3b84 | 763 | break; |
066a3b5b | 764 | |
4eaf3b84 ED |
765 | if (sch->flags & TCQ_F_NOPARENT) |
766 | break; | |
767 | /* TODO: perform the search on a per txq basis */ | |
5ce2d488 | 768 | sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); |
ffc8fefa | 769 | if (sch == NULL) { |
4eaf3b84 ED |
770 | WARN_ON_ONCE(parentid != TC_H_ROOT); |
771 | break; | |
ffc8fefa | 772 | } |
43effa1e PM |
773 | cops = sch->ops->cl_ops; |
774 | if (cops->qlen_notify) { | |
775 | cl = cops->get(sch, parentid); | |
776 | cops->qlen_notify(sch, cl); | |
777 | cops->put(sch, cl); | |
778 | } | |
779 | sch->q.qlen -= n; | |
2ccccf5f | 780 | sch->qstats.backlog -= len; |
25331d6c | 781 | __qdisc_qstats_drop(sch, drops); |
43effa1e | 782 | } |
4eaf3b84 | 783 | rcu_read_unlock(); |
43effa1e | 784 | } |
2ccccf5f | 785 | EXPORT_SYMBOL(qdisc_tree_reduce_backlog); |
1da177e4 | 786 | |
7316ae88 TG |
787 | static void notify_and_destroy(struct net *net, struct sk_buff *skb, |
788 | struct nlmsghdr *n, u32 clid, | |
99194cff DM |
789 | struct Qdisc *old, struct Qdisc *new) |
790 | { | |
791 | if (new || old) | |
7316ae88 | 792 | qdisc_notify(net, skb, n, clid, old, new); |
1da177e4 | 793 | |
4d8863a2 | 794 | if (old) |
99194cff | 795 | qdisc_destroy(old); |
99194cff DM |
796 | } |
797 | ||
798 | /* Graft qdisc "new" to class "classid" of qdisc "parent" or | |
799 | * to device "dev". | |
800 | * | |
801 | * When appropriate send a netlink notification using 'skb' | |
802 | * and "n". | |
803 | * | |
804 | * On success, destroy old qdisc. | |
1da177e4 LT |
805 | */ |
806 | ||
807 | static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, | |
99194cff DM |
808 | struct sk_buff *skb, struct nlmsghdr *n, u32 classid, |
809 | struct Qdisc *new, struct Qdisc *old) | |
1da177e4 | 810 | { |
99194cff | 811 | struct Qdisc *q = old; |
7316ae88 | 812 | struct net *net = dev_net(dev); |
1da177e4 | 813 | int err = 0; |
1da177e4 | 814 | |
10297b99 | 815 | if (parent == NULL) { |
99194cff DM |
816 | unsigned int i, num_q, ingress; |
817 | ||
818 | ingress = 0; | |
819 | num_q = dev->num_tx_queues; | |
8d50b53d DM |
820 | if ((q && q->flags & TCQ_F_INGRESS) || |
821 | (new && new->flags & TCQ_F_INGRESS)) { | |
99194cff DM |
822 | num_q = 1; |
823 | ingress = 1; | |
24824a09 ED |
824 | if (!dev_ingress_queue(dev)) |
825 | return -ENOENT; | |
99194cff DM |
826 | } |
827 | ||
828 | if (dev->flags & IFF_UP) | |
829 | dev_deactivate(dev); | |
830 | ||
86e363dc WC |
831 | if (new && new->ops->attach) |
832 | goto skip; | |
6ec1c69a | 833 | |
99194cff | 834 | for (i = 0; i < num_q; i++) { |
24824a09 | 835 | struct netdev_queue *dev_queue = dev_ingress_queue(dev); |
99194cff DM |
836 | |
837 | if (!ingress) | |
838 | dev_queue = netdev_get_tx_queue(dev, i); | |
839 | ||
8d50b53d DM |
840 | old = dev_graft_qdisc(dev_queue, new); |
841 | if (new && i > 0) | |
842 | atomic_inc(&new->refcnt); | |
843 | ||
036d6a67 JP |
844 | if (!ingress) |
845 | qdisc_destroy(old); | |
1da177e4 | 846 | } |
99194cff | 847 | |
86e363dc | 848 | skip: |
036d6a67 | 849 | if (!ingress) { |
7316ae88 TG |
850 | notify_and_destroy(net, skb, n, classid, |
851 | dev->qdisc, new); | |
036d6a67 JP |
852 | if (new && !new->ops->attach) |
853 | atomic_inc(&new->refcnt); | |
854 | dev->qdisc = new ? : &noop_qdisc; | |
86e363dc WC |
855 | |
856 | if (new && new->ops->attach) | |
857 | new->ops->attach(new); | |
036d6a67 | 858 | } else { |
7316ae88 | 859 | notify_and_destroy(net, skb, n, classid, old, new); |
036d6a67 | 860 | } |
af356afa | 861 | |
99194cff DM |
862 | if (dev->flags & IFF_UP) |
863 | dev_activate(dev); | |
1da177e4 | 864 | } else { |
20fea08b | 865 | const struct Qdisc_class_ops *cops = parent->ops->cl_ops; |
1da177e4 | 866 | |
c9f1d038 PM |
867 | err = -EOPNOTSUPP; |
868 | if (cops && cops->graft) { | |
1da177e4 LT |
869 | unsigned long cl = cops->get(parent, classid); |
870 | if (cl) { | |
99194cff | 871 | err = cops->graft(parent, cl, new, &old); |
1da177e4 | 872 | cops->put(parent, cl); |
c9f1d038 PM |
873 | } else |
874 | err = -ENOENT; | |
1da177e4 | 875 | } |
99194cff | 876 | if (!err) |
7316ae88 | 877 | notify_and_destroy(net, skb, n, classid, old, new); |
1da177e4 LT |
878 | } |
879 | return err; | |
880 | } | |
881 | ||
25bfcd5a JP |
882 | /* lockdep annotation is needed for ingress; egress gets it only for name */ |
883 | static struct lock_class_key qdisc_tx_lock; | |
884 | static struct lock_class_key qdisc_rx_lock; | |
885 | ||
1da177e4 LT |
886 | /* |
887 | Allocate and initialize new qdisc. | |
888 | ||
889 | Parameters are passed via opt. | |
890 | */ | |
891 | ||
5a7a5555 JHS |
892 | static struct Qdisc *qdisc_create(struct net_device *dev, |
893 | struct netdev_queue *dev_queue, | |
894 | struct Qdisc *p, u32 parent, u32 handle, | |
895 | struct nlattr **tca, int *errp) | |
1da177e4 LT |
896 | { |
897 | int err; | |
1e90474c | 898 | struct nlattr *kind = tca[TCA_KIND]; |
1da177e4 LT |
899 | struct Qdisc *sch; |
900 | struct Qdisc_ops *ops; | |
175f9c1b | 901 | struct qdisc_size_table *stab; |
1da177e4 LT |
902 | |
903 | ops = qdisc_lookup_ops(kind); | |
95a5afca | 904 | #ifdef CONFIG_MODULES |
1da177e4 LT |
905 | if (ops == NULL && kind != NULL) { |
906 | char name[IFNAMSIZ]; | |
1e90474c | 907 | if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { |
1da177e4 LT |
908 | /* We dropped the RTNL semaphore in order to |
909 | * perform the module load. So, even if we | |
910 | * succeeded in loading the module we have to | |
911 | * tell the caller to replay the request. We | |
912 | * indicate this using -EAGAIN. | |
913 | * We replay the request because the device may | |
914 | * go away in the mean time. | |
915 | */ | |
916 | rtnl_unlock(); | |
917 | request_module("sch_%s", name); | |
918 | rtnl_lock(); | |
919 | ops = qdisc_lookup_ops(kind); | |
920 | if (ops != NULL) { | |
921 | /* We will try again qdisc_lookup_ops, | |
922 | * so don't keep a reference. | |
923 | */ | |
924 | module_put(ops->owner); | |
925 | err = -EAGAIN; | |
926 | goto err_out; | |
927 | } | |
928 | } | |
929 | } | |
930 | #endif | |
931 | ||
b9e2cc0f | 932 | err = -ENOENT; |
1da177e4 LT |
933 | if (ops == NULL) |
934 | goto err_out; | |
935 | ||
5ce2d488 | 936 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
937 | if (IS_ERR(sch)) { |
938 | err = PTR_ERR(sch); | |
1da177e4 | 939 | goto err_out2; |
3d54b82f | 940 | } |
1da177e4 | 941 | |
ffc8fefa PM |
942 | sch->parent = parent; |
943 | ||
3d54b82f | 944 | if (handle == TC_H_INGRESS) { |
1da177e4 | 945 | sch->flags |= TCQ_F_INGRESS; |
3d54b82f | 946 | handle = TC_H_MAKE(TC_H_INGRESS, 0); |
25bfcd5a | 947 | lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); |
fd44de7c | 948 | } else { |
fd44de7c PM |
949 | if (handle == 0) { |
950 | handle = qdisc_alloc_handle(dev); | |
951 | err = -ENOMEM; | |
952 | if (handle == 0) | |
953 | goto err_out3; | |
954 | } | |
25bfcd5a | 955 | lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); |
1abbe139 | 956 | if (!netif_is_multiqueue(dev)) |
225734de | 957 | sch->flags |= TCQ_F_ONETXQUEUE; |
1da177e4 LT |
958 | } |
959 | ||
3d54b82f | 960 | sch->handle = handle; |
1da177e4 | 961 | |
84c46dd8 JDB |
962 | /* This exist to keep backward compatible with a userspace |
963 | * loophole, what allowed userspace to get IFF_NO_QUEUE | |
964 | * facility on older kernels by setting tx_queue_len=0 (prior | |
965 | * to qdisc init), and then forgot to reinit tx_queue_len | |
966 | * before again attaching a qdisc. | |
967 | */ | |
968 | if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) { | |
969 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; | |
970 | netdev_info(dev, "Caught tx_queue_len zero misconfig\n"); | |
971 | } | |
972 | ||
1e90474c | 973 | if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { |
22e0f8b9 JF |
974 | if (qdisc_is_percpu_stats(sch)) { |
975 | sch->cpu_bstats = | |
7c1c97d5 | 976 | netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu); |
22e0f8b9 JF |
977 | if (!sch->cpu_bstats) |
978 | goto err_out4; | |
b0ab6f92 JF |
979 | |
980 | sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue); | |
981 | if (!sch->cpu_qstats) | |
982 | goto err_out4; | |
22e0f8b9 JF |
983 | } |
984 | ||
175f9c1b JK |
985 | if (tca[TCA_STAB]) { |
986 | stab = qdisc_get_stab(tca[TCA_STAB]); | |
987 | if (IS_ERR(stab)) { | |
988 | err = PTR_ERR(stab); | |
7c64b9f3 | 989 | goto err_out4; |
175f9c1b | 990 | } |
a2da570d | 991 | rcu_assign_pointer(sch->stab, stab); |
175f9c1b | 992 | } |
1e90474c | 993 | if (tca[TCA_RATE]) { |
edb09eb1 | 994 | seqcount_t *running; |
f6f9b93f | 995 | |
23bcf634 PM |
996 | err = -EOPNOTSUPP; |
997 | if (sch->flags & TCQ_F_MQROOT) | |
998 | goto err_out4; | |
999 | ||
f6f9b93f | 1000 | if ((sch->parent != TC_H_ROOT) && |
23bcf634 PM |
1001 | !(sch->flags & TCQ_F_INGRESS) && |
1002 | (!p || !(p->flags & TCQ_F_MQROOT))) | |
edb09eb1 | 1003 | running = qdisc_root_sleeping_running(sch); |
f6f9b93f | 1004 | else |
edb09eb1 | 1005 | running = &sch->running; |
f6f9b93f | 1006 | |
22e0f8b9 JF |
1007 | err = gen_new_estimator(&sch->bstats, |
1008 | sch->cpu_bstats, | |
1009 | &sch->rate_est, | |
edb09eb1 ED |
1010 | NULL, |
1011 | running, | |
22e0f8b9 | 1012 | tca[TCA_RATE]); |
23bcf634 PM |
1013 | if (err) |
1014 | goto err_out4; | |
023e09a7 | 1015 | } |
f6e0b239 | 1016 | |
49b49971 | 1017 | qdisc_hash_add(sch, false); |
1da177e4 | 1018 | |
1da177e4 LT |
1019 | return sch; |
1020 | } | |
87b60cfa ED |
1021 | /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */ |
1022 | ops->destroy(sch); | |
1da177e4 LT |
1023 | err_out3: |
1024 | dev_put(dev); | |
3d54b82f | 1025 | kfree((char *) sch - sch->padded); |
1da177e4 LT |
1026 | err_out2: |
1027 | module_put(ops->owner); | |
1028 | err_out: | |
1029 | *errp = err; | |
1da177e4 | 1030 | return NULL; |
23bcf634 PM |
1031 | |
1032 | err_out4: | |
22e0f8b9 | 1033 | free_percpu(sch->cpu_bstats); |
b0ab6f92 | 1034 | free_percpu(sch->cpu_qstats); |
23bcf634 PM |
1035 | /* |
1036 | * Any broken qdiscs that would require a ops->reset() here? | |
1037 | * The qdisc was never in action so it shouldn't be necessary. | |
1038 | */ | |
a2da570d | 1039 | qdisc_put_stab(rtnl_dereference(sch->stab)); |
23bcf634 PM |
1040 | if (ops->destroy) |
1041 | ops->destroy(sch); | |
1042 | goto err_out3; | |
1da177e4 LT |
1043 | } |
1044 | ||
1e90474c | 1045 | static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) |
1da177e4 | 1046 | { |
a2da570d | 1047 | struct qdisc_size_table *ostab, *stab = NULL; |
175f9c1b | 1048 | int err = 0; |
1da177e4 | 1049 | |
175f9c1b | 1050 | if (tca[TCA_OPTIONS]) { |
1da177e4 LT |
1051 | if (sch->ops->change == NULL) |
1052 | return -EINVAL; | |
1e90474c | 1053 | err = sch->ops->change(sch, tca[TCA_OPTIONS]); |
1da177e4 LT |
1054 | if (err) |
1055 | return err; | |
1056 | } | |
175f9c1b JK |
1057 | |
1058 | if (tca[TCA_STAB]) { | |
1059 | stab = qdisc_get_stab(tca[TCA_STAB]); | |
1060 | if (IS_ERR(stab)) | |
1061 | return PTR_ERR(stab); | |
1062 | } | |
1063 | ||
a2da570d ED |
1064 | ostab = rtnl_dereference(sch->stab); |
1065 | rcu_assign_pointer(sch->stab, stab); | |
1066 | qdisc_put_stab(ostab); | |
175f9c1b | 1067 | |
23bcf634 | 1068 | if (tca[TCA_RATE]) { |
71bcb09a SH |
1069 | /* NB: ignores errors from replace_estimator |
1070 | because change can't be undone. */ | |
23bcf634 PM |
1071 | if (sch->flags & TCQ_F_MQROOT) |
1072 | goto out; | |
22e0f8b9 JF |
1073 | gen_replace_estimator(&sch->bstats, |
1074 | sch->cpu_bstats, | |
1075 | &sch->rate_est, | |
edb09eb1 ED |
1076 | NULL, |
1077 | qdisc_root_sleeping_running(sch), | |
22e0f8b9 | 1078 | tca[TCA_RATE]); |
23bcf634 PM |
1079 | } |
1080 | out: | |
1da177e4 LT |
1081 | return 0; |
1082 | } | |
1083 | ||
cc7ec456 ED |
1084 | struct check_loop_arg { |
1085 | struct qdisc_walker w; | |
1da177e4 LT |
1086 | struct Qdisc *p; |
1087 | int depth; | |
1088 | }; | |
1089 | ||
5a7a5555 JHS |
1090 | static int check_loop_fn(struct Qdisc *q, unsigned long cl, |
1091 | struct qdisc_walker *w); | |
1da177e4 LT |
1092 | |
1093 | static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) | |
1094 | { | |
1095 | struct check_loop_arg arg; | |
1096 | ||
1097 | if (q->ops->cl_ops == NULL) | |
1098 | return 0; | |
1099 | ||
1100 | arg.w.stop = arg.w.skip = arg.w.count = 0; | |
1101 | arg.w.fn = check_loop_fn; | |
1102 | arg.depth = depth; | |
1103 | arg.p = p; | |
1104 | q->ops->cl_ops->walk(q, &arg.w); | |
1105 | return arg.w.stop ? -ELOOP : 0; | |
1106 | } | |
1107 | ||
1108 | static int | |
1109 | check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) | |
1110 | { | |
1111 | struct Qdisc *leaf; | |
20fea08b | 1112 | const struct Qdisc_class_ops *cops = q->ops->cl_ops; |
1da177e4 LT |
1113 | struct check_loop_arg *arg = (struct check_loop_arg *)w; |
1114 | ||
1115 | leaf = cops->leaf(q, cl); | |
1116 | if (leaf) { | |
1117 | if (leaf == arg->p || arg->depth > 7) | |
1118 | return -ELOOP; | |
1119 | return check_loop(leaf, arg->p, arg->depth + 1); | |
1120 | } | |
1121 | return 0; | |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * Delete/get qdisc. | |
1126 | */ | |
1127 | ||
c21ef3e3 DA |
1128 | static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, |
1129 | struct netlink_ext_ack *extack) | |
1da177e4 | 1130 | { |
3b1e0a65 | 1131 | struct net *net = sock_net(skb->sk); |
02ef22ca | 1132 | struct tcmsg *tcm = nlmsg_data(n); |
1e90474c | 1133 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 | 1134 | struct net_device *dev; |
de179c8c | 1135 | u32 clid; |
1da177e4 LT |
1136 | struct Qdisc *q = NULL; |
1137 | struct Qdisc *p = NULL; | |
1138 | int err; | |
1139 | ||
4e8bbb81 | 1140 | if ((n->nlmsg_type != RTM_GETQDISC) && |
5f013c9b | 1141 | !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) |
dfc47ef8 EB |
1142 | return -EPERM; |
1143 | ||
c21ef3e3 | 1144 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); |
1e90474c PM |
1145 | if (err < 0) |
1146 | return err; | |
1147 | ||
de179c8c H |
1148 | dev = __dev_get_by_index(net, tcm->tcm_ifindex); |
1149 | if (!dev) | |
1150 | return -ENODEV; | |
1151 | ||
1152 | clid = tcm->tcm_parent; | |
1da177e4 LT |
1153 | if (clid) { |
1154 | if (clid != TC_H_ROOT) { | |
1155 | if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { | |
cc7ec456 ED |
1156 | p = qdisc_lookup(dev, TC_H_MAJ(clid)); |
1157 | if (!p) | |
1da177e4 LT |
1158 | return -ENOENT; |
1159 | q = qdisc_leaf(p, clid); | |
cc7ec456 ED |
1160 | } else if (dev_ingress_queue(dev)) { |
1161 | q = dev_ingress_queue(dev)->qdisc_sleeping; | |
10297b99 | 1162 | } |
1da177e4 | 1163 | } else { |
af356afa | 1164 | q = dev->qdisc; |
1da177e4 LT |
1165 | } |
1166 | if (!q) | |
1167 | return -ENOENT; | |
1168 | ||
1169 | if (tcm->tcm_handle && q->handle != tcm->tcm_handle) | |
1170 | return -EINVAL; | |
1171 | } else { | |
cc7ec456 ED |
1172 | q = qdisc_lookup(dev, tcm->tcm_handle); |
1173 | if (!q) | |
1da177e4 LT |
1174 | return -ENOENT; |
1175 | } | |
1176 | ||
1e90474c | 1177 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
1178 | return -EINVAL; |
1179 | ||
1180 | if (n->nlmsg_type == RTM_DELQDISC) { | |
1181 | if (!clid) | |
1182 | return -EINVAL; | |
1183 | if (q->handle == 0) | |
1184 | return -ENOENT; | |
cc7ec456 ED |
1185 | err = qdisc_graft(dev, p, skb, n, clid, NULL, q); |
1186 | if (err != 0) | |
1da177e4 | 1187 | return err; |
1da177e4 | 1188 | } else { |
7316ae88 | 1189 | qdisc_notify(net, skb, n, clid, NULL, q); |
1da177e4 LT |
1190 | } |
1191 | return 0; | |
1192 | } | |
1193 | ||
1194 | /* | |
cc7ec456 | 1195 | * Create/change qdisc. |
1da177e4 LT |
1196 | */ |
1197 | ||
c21ef3e3 DA |
1198 | static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, |
1199 | struct netlink_ext_ack *extack) | |
1da177e4 | 1200 | { |
3b1e0a65 | 1201 | struct net *net = sock_net(skb->sk); |
1da177e4 | 1202 | struct tcmsg *tcm; |
1e90474c | 1203 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
1204 | struct net_device *dev; |
1205 | u32 clid; | |
1206 | struct Qdisc *q, *p; | |
1207 | int err; | |
1208 | ||
5f013c9b | 1209 | if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) |
dfc47ef8 EB |
1210 | return -EPERM; |
1211 | ||
1da177e4 LT |
1212 | replay: |
1213 | /* Reinit, just in case something touches this. */ | |
c21ef3e3 | 1214 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); |
de179c8c H |
1215 | if (err < 0) |
1216 | return err; | |
1217 | ||
02ef22ca | 1218 | tcm = nlmsg_data(n); |
1da177e4 LT |
1219 | clid = tcm->tcm_parent; |
1220 | q = p = NULL; | |
1221 | ||
cc7ec456 ED |
1222 | dev = __dev_get_by_index(net, tcm->tcm_ifindex); |
1223 | if (!dev) | |
1da177e4 LT |
1224 | return -ENODEV; |
1225 | ||
1e90474c | 1226 | |
1da177e4 LT |
1227 | if (clid) { |
1228 | if (clid != TC_H_ROOT) { | |
1229 | if (clid != TC_H_INGRESS) { | |
cc7ec456 ED |
1230 | p = qdisc_lookup(dev, TC_H_MAJ(clid)); |
1231 | if (!p) | |
1da177e4 LT |
1232 | return -ENOENT; |
1233 | q = qdisc_leaf(p, clid); | |
cc7ec456 ED |
1234 | } else if (dev_ingress_queue_create(dev)) { |
1235 | q = dev_ingress_queue(dev)->qdisc_sleeping; | |
1da177e4 LT |
1236 | } |
1237 | } else { | |
af356afa | 1238 | q = dev->qdisc; |
1da177e4 LT |
1239 | } |
1240 | ||
1241 | /* It may be default qdisc, ignore it */ | |
1242 | if (q && q->handle == 0) | |
1243 | q = NULL; | |
1244 | ||
1245 | if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { | |
1246 | if (tcm->tcm_handle) { | |
cc7ec456 | 1247 | if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) |
1da177e4 LT |
1248 | return -EEXIST; |
1249 | if (TC_H_MIN(tcm->tcm_handle)) | |
1250 | return -EINVAL; | |
cc7ec456 ED |
1251 | q = qdisc_lookup(dev, tcm->tcm_handle); |
1252 | if (!q) | |
1da177e4 | 1253 | goto create_n_graft; |
cc7ec456 | 1254 | if (n->nlmsg_flags & NLM_F_EXCL) |
1da177e4 | 1255 | return -EEXIST; |
1e90474c | 1256 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
1257 | return -EINVAL; |
1258 | if (q == p || | |
1259 | (p && check_loop(q, p, 0))) | |
1260 | return -ELOOP; | |
1261 | atomic_inc(&q->refcnt); | |
1262 | goto graft; | |
1263 | } else { | |
cc7ec456 | 1264 | if (!q) |
1da177e4 LT |
1265 | goto create_n_graft; |
1266 | ||
1267 | /* This magic test requires explanation. | |
1268 | * | |
1269 | * We know, that some child q is already | |
1270 | * attached to this parent and have choice: | |
1271 | * either to change it or to create/graft new one. | |
1272 | * | |
1273 | * 1. We are allowed to create/graft only | |
1274 | * if CREATE and REPLACE flags are set. | |
1275 | * | |
1276 | * 2. If EXCL is set, requestor wanted to say, | |
1277 | * that qdisc tcm_handle is not expected | |
1278 | * to exist, so that we choose create/graft too. | |
1279 | * | |
1280 | * 3. The last case is when no flags are set. | |
1281 | * Alas, it is sort of hole in API, we | |
1282 | * cannot decide what to do unambiguously. | |
1283 | * For now we select create/graft, if | |
1284 | * user gave KIND, which does not match existing. | |
1285 | */ | |
cc7ec456 ED |
1286 | if ((n->nlmsg_flags & NLM_F_CREATE) && |
1287 | (n->nlmsg_flags & NLM_F_REPLACE) && | |
1288 | ((n->nlmsg_flags & NLM_F_EXCL) || | |
1e90474c PM |
1289 | (tca[TCA_KIND] && |
1290 | nla_strcmp(tca[TCA_KIND], q->ops->id)))) | |
1da177e4 LT |
1291 | goto create_n_graft; |
1292 | } | |
1293 | } | |
1294 | } else { | |
1295 | if (!tcm->tcm_handle) | |
1296 | return -EINVAL; | |
1297 | q = qdisc_lookup(dev, tcm->tcm_handle); | |
1298 | } | |
1299 | ||
1300 | /* Change qdisc parameters */ | |
1301 | if (q == NULL) | |
1302 | return -ENOENT; | |
cc7ec456 | 1303 | if (n->nlmsg_flags & NLM_F_EXCL) |
1da177e4 | 1304 | return -EEXIST; |
1e90474c | 1305 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) |
1da177e4 LT |
1306 | return -EINVAL; |
1307 | err = qdisc_change(q, tca); | |
1308 | if (err == 0) | |
7316ae88 | 1309 | qdisc_notify(net, skb, n, clid, NULL, q); |
1da177e4 LT |
1310 | return err; |
1311 | ||
1312 | create_n_graft: | |
cc7ec456 | 1313 | if (!(n->nlmsg_flags & NLM_F_CREATE)) |
1da177e4 | 1314 | return -ENOENT; |
24824a09 ED |
1315 | if (clid == TC_H_INGRESS) { |
1316 | if (dev_ingress_queue(dev)) | |
1317 | q = qdisc_create(dev, dev_ingress_queue(dev), p, | |
1318 | tcm->tcm_parent, tcm->tcm_parent, | |
1319 | tca, &err); | |
1320 | else | |
1321 | err = -ENOENT; | |
1322 | } else { | |
926e61b7 | 1323 | struct netdev_queue *dev_queue; |
6ec1c69a DM |
1324 | |
1325 | if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) | |
926e61b7 JP |
1326 | dev_queue = p->ops->cl_ops->select_queue(p, tcm); |
1327 | else if (p) | |
1328 | dev_queue = p->dev_queue; | |
1329 | else | |
1330 | dev_queue = netdev_get_tx_queue(dev, 0); | |
6ec1c69a | 1331 | |
926e61b7 | 1332 | q = qdisc_create(dev, dev_queue, p, |
bb949fbd | 1333 | tcm->tcm_parent, tcm->tcm_handle, |
ffc8fefa | 1334 | tca, &err); |
6ec1c69a | 1335 | } |
1da177e4 LT |
1336 | if (q == NULL) { |
1337 | if (err == -EAGAIN) | |
1338 | goto replay; | |
1339 | return err; | |
1340 | } | |
1341 | ||
1342 | graft: | |
e5befbd9 IJ |
1343 | err = qdisc_graft(dev, p, skb, n, clid, q, NULL); |
1344 | if (err) { | |
1345 | if (q) | |
1346 | qdisc_destroy(q); | |
1347 | return err; | |
1da177e4 | 1348 | } |
e5befbd9 | 1349 | |
1da177e4 LT |
1350 | return 0; |
1351 | } | |
1352 | ||
1353 | static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, | |
15e47304 | 1354 | u32 portid, u32 seq, u16 flags, int event) |
1da177e4 | 1355 | { |
22e0f8b9 | 1356 | struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL; |
b0ab6f92 | 1357 | struct gnet_stats_queue __percpu *cpu_qstats = NULL; |
1da177e4 LT |
1358 | struct tcmsg *tcm; |
1359 | struct nlmsghdr *nlh; | |
27a884dc | 1360 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 | 1361 | struct gnet_dump d; |
a2da570d | 1362 | struct qdisc_size_table *stab; |
64015853 | 1363 | __u32 qlen; |
1da177e4 | 1364 | |
fba373d2 | 1365 | cond_resched(); |
15e47304 | 1366 | nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); |
02ef22ca DM |
1367 | if (!nlh) |
1368 | goto out_nlmsg_trim; | |
1369 | tcm = nlmsg_data(nlh); | |
1da177e4 | 1370 | tcm->tcm_family = AF_UNSPEC; |
9ef1d4c7 PM |
1371 | tcm->tcm__pad1 = 0; |
1372 | tcm->tcm__pad2 = 0; | |
5ce2d488 | 1373 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex; |
1da177e4 LT |
1374 | tcm->tcm_parent = clid; |
1375 | tcm->tcm_handle = q->handle; | |
1376 | tcm->tcm_info = atomic_read(&q->refcnt); | |
1b34ec43 DM |
1377 | if (nla_put_string(skb, TCA_KIND, q->ops->id)) |
1378 | goto nla_put_failure; | |
1da177e4 | 1379 | if (q->ops->dump && q->ops->dump(q, skb) < 0) |
1e90474c | 1380 | goto nla_put_failure; |
64015853 | 1381 | qlen = q->q.qlen; |
1da177e4 | 1382 | |
a2da570d ED |
1383 | stab = rtnl_dereference(q->stab); |
1384 | if (stab && qdisc_dump_stab(skb, stab) < 0) | |
175f9c1b JK |
1385 | goto nla_put_failure; |
1386 | ||
102396ae | 1387 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, |
edb09eb1 | 1388 | NULL, &d, TCA_PAD) < 0) |
1e90474c | 1389 | goto nla_put_failure; |
1da177e4 LT |
1390 | |
1391 | if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) | |
1e90474c | 1392 | goto nla_put_failure; |
1da177e4 | 1393 | |
b0ab6f92 | 1394 | if (qdisc_is_percpu_stats(q)) { |
22e0f8b9 | 1395 | cpu_bstats = q->cpu_bstats; |
b0ab6f92 JF |
1396 | cpu_qstats = q->cpu_qstats; |
1397 | } | |
22e0f8b9 | 1398 | |
edb09eb1 ED |
1399 | if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q), |
1400 | &d, cpu_bstats, &q->bstats) < 0 || | |
1c0d32fd | 1401 | gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || |
b0ab6f92 | 1402 | gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0) |
1e90474c | 1403 | goto nla_put_failure; |
10297b99 | 1404 | |
1da177e4 | 1405 | if (gnet_stats_finish_copy(&d) < 0) |
1e90474c | 1406 | goto nla_put_failure; |
10297b99 | 1407 | |
27a884dc | 1408 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
1409 | return skb->len; |
1410 | ||
02ef22ca | 1411 | out_nlmsg_trim: |
1e90474c | 1412 | nla_put_failure: |
dc5fc579 | 1413 | nlmsg_trim(skb, b); |
1da177e4 LT |
1414 | return -1; |
1415 | } | |
1416 | ||
49b49971 | 1417 | static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible) |
53b0f080 | 1418 | { |
49b49971 JK |
1419 | if (q->flags & TCQ_F_BUILTIN) |
1420 | return true; | |
1421 | if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible) | |
1422 | return true; | |
1423 | ||
1424 | return false; | |
53b0f080 ED |
1425 | } |
1426 | ||
7316ae88 TG |
1427 | static int qdisc_notify(struct net *net, struct sk_buff *oskb, |
1428 | struct nlmsghdr *n, u32 clid, | |
1429 | struct Qdisc *old, struct Qdisc *new) | |
1da177e4 LT |
1430 | { |
1431 | struct sk_buff *skb; | |
15e47304 | 1432 | u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; |
1da177e4 LT |
1433 | |
1434 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1435 | if (!skb) | |
1436 | return -ENOBUFS; | |
1437 | ||
49b49971 | 1438 | if (old && !tc_qdisc_dump_ignore(old, false)) { |
15e47304 | 1439 | if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq, |
cc7ec456 | 1440 | 0, RTM_DELQDISC) < 0) |
1da177e4 LT |
1441 | goto err_out; |
1442 | } | |
49b49971 | 1443 | if (new && !tc_qdisc_dump_ignore(new, false)) { |
15e47304 | 1444 | if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq, |
cc7ec456 | 1445 | old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) |
1da177e4 LT |
1446 | goto err_out; |
1447 | } | |
1448 | ||
1449 | if (skb->len) | |
15e47304 | 1450 | return rtnetlink_send(skb, net, portid, RTNLGRP_TC, |
cc7ec456 | 1451 | n->nlmsg_flags & NLM_F_ECHO); |
1da177e4 LT |
1452 | |
1453 | err_out: | |
1454 | kfree_skb(skb); | |
1455 | return -EINVAL; | |
1456 | } | |
1457 | ||
30723673 DM |
1458 | static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, |
1459 | struct netlink_callback *cb, | |
49b49971 JK |
1460 | int *q_idx_p, int s_q_idx, bool recur, |
1461 | bool dump_invisible) | |
30723673 DM |
1462 | { |
1463 | int ret = 0, q_idx = *q_idx_p; | |
1464 | struct Qdisc *q; | |
59cc1f61 | 1465 | int b; |
30723673 DM |
1466 | |
1467 | if (!root) | |
1468 | return 0; | |
1469 | ||
1470 | q = root; | |
1471 | if (q_idx < s_q_idx) { | |
1472 | q_idx++; | |
1473 | } else { | |
49b49971 | 1474 | if (!tc_qdisc_dump_ignore(q, dump_invisible) && |
15e47304 | 1475 | tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, |
5a7a5555 JHS |
1476 | cb->nlh->nlmsg_seq, NLM_F_MULTI, |
1477 | RTM_NEWQDISC) <= 0) | |
30723673 DM |
1478 | goto done; |
1479 | q_idx++; | |
1480 | } | |
69012ae4 | 1481 | |
ea327469 JK |
1482 | /* If dumping singletons, there is no qdisc_dev(root) and the singleton |
1483 | * itself has already been dumped. | |
1484 | * | |
1485 | * If we've already dumped the top-level (ingress) qdisc above and the global | |
1486 | * qdisc hashtable, we don't want to hit it again | |
1487 | */ | |
1488 | if (!qdisc_dev(root) || !recur) | |
69012ae4 JK |
1489 | goto out; |
1490 | ||
59cc1f61 | 1491 | hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { |
30723673 DM |
1492 | if (q_idx < s_q_idx) { |
1493 | q_idx++; | |
1494 | continue; | |
1495 | } | |
49b49971 | 1496 | if (!tc_qdisc_dump_ignore(q, dump_invisible) && |
15e47304 | 1497 | tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, |
5a7a5555 JHS |
1498 | cb->nlh->nlmsg_seq, NLM_F_MULTI, |
1499 | RTM_NEWQDISC) <= 0) | |
30723673 DM |
1500 | goto done; |
1501 | q_idx++; | |
1502 | } | |
1503 | ||
1504 | out: | |
1505 | *q_idx_p = q_idx; | |
1506 | return ret; | |
1507 | done: | |
1508 | ret = -1; | |
1509 | goto out; | |
1510 | } | |
1511 | ||
1da177e4 LT |
1512 | static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) |
1513 | { | |
3b1e0a65 | 1514 | struct net *net = sock_net(skb->sk); |
1da177e4 LT |
1515 | int idx, q_idx; |
1516 | int s_idx, s_q_idx; | |
1517 | struct net_device *dev; | |
49b49971 JK |
1518 | const struct nlmsghdr *nlh = cb->nlh; |
1519 | struct tcmsg *tcm = nlmsg_data(nlh); | |
1520 | struct nlattr *tca[TCA_MAX + 1]; | |
1521 | int err; | |
1da177e4 LT |
1522 | |
1523 | s_idx = cb->args[0]; | |
1524 | s_q_idx = q_idx = cb->args[1]; | |
f1e9016d | 1525 | |
7562f876 | 1526 | idx = 0; |
15dc36eb | 1527 | ASSERT_RTNL(); |
49b49971 | 1528 | |
fceb6435 | 1529 | err = nlmsg_parse(nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL); |
49b49971 JK |
1530 | if (err < 0) |
1531 | return err; | |
1532 | ||
15dc36eb | 1533 | for_each_netdev(net, dev) { |
30723673 DM |
1534 | struct netdev_queue *dev_queue; |
1535 | ||
1da177e4 | 1536 | if (idx < s_idx) |
7562f876 | 1537 | goto cont; |
1da177e4 LT |
1538 | if (idx > s_idx) |
1539 | s_q_idx = 0; | |
1da177e4 | 1540 | q_idx = 0; |
30723673 | 1541 | |
5a7a5555 | 1542 | if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx, |
49b49971 | 1543 | true, tca[TCA_DUMP_INVISIBLE]) < 0) |
30723673 DM |
1544 | goto done; |
1545 | ||
24824a09 ED |
1546 | dev_queue = dev_ingress_queue(dev); |
1547 | if (dev_queue && | |
1548 | tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, | |
49b49971 JK |
1549 | &q_idx, s_q_idx, false, |
1550 | tca[TCA_DUMP_INVISIBLE]) < 0) | |
30723673 DM |
1551 | goto done; |
1552 | ||
7562f876 PE |
1553 | cont: |
1554 | idx++; | |
1da177e4 LT |
1555 | } |
1556 | ||
1557 | done: | |
1da177e4 LT |
1558 | cb->args[0] = idx; |
1559 | cb->args[1] = q_idx; | |
1560 | ||
1561 | return skb->len; | |
1562 | } | |
1563 | ||
1564 | ||
1565 | ||
1566 | /************************************************ | |
1567 | * Traffic classes manipulation. * | |
1568 | ************************************************/ | |
1569 | ||
1570 | ||
1571 | ||
c21ef3e3 DA |
1572 | static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, |
1573 | struct netlink_ext_ack *extack) | |
1da177e4 | 1574 | { |
3b1e0a65 | 1575 | struct net *net = sock_net(skb->sk); |
02ef22ca | 1576 | struct tcmsg *tcm = nlmsg_data(n); |
1e90474c | 1577 | struct nlattr *tca[TCA_MAX + 1]; |
1da177e4 LT |
1578 | struct net_device *dev; |
1579 | struct Qdisc *q = NULL; | |
20fea08b | 1580 | const struct Qdisc_class_ops *cops; |
1da177e4 LT |
1581 | unsigned long cl = 0; |
1582 | unsigned long new_cl; | |
de179c8c H |
1583 | u32 portid; |
1584 | u32 clid; | |
1585 | u32 qid; | |
1da177e4 LT |
1586 | int err; |
1587 | ||
4e8bbb81 | 1588 | if ((n->nlmsg_type != RTM_GETTCLASS) && |
5f013c9b | 1589 | !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) |
dfc47ef8 EB |
1590 | return -EPERM; |
1591 | ||
c21ef3e3 | 1592 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, extack); |
1e90474c PM |
1593 | if (err < 0) |
1594 | return err; | |
1595 | ||
de179c8c H |
1596 | dev = __dev_get_by_index(net, tcm->tcm_ifindex); |
1597 | if (!dev) | |
1598 | return -ENODEV; | |
1599 | ||
1da177e4 LT |
1600 | /* |
1601 | parent == TC_H_UNSPEC - unspecified parent. | |
1602 | parent == TC_H_ROOT - class is root, which has no parent. | |
1603 | parent == X:0 - parent is root class. | |
1604 | parent == X:Y - parent is a node in hierarchy. | |
1605 | parent == 0:Y - parent is X:Y, where X:0 is qdisc. | |
1606 | ||
1607 | handle == 0:0 - generate handle from kernel pool. | |
1608 | handle == 0:Y - class is X:Y, where X:0 is qdisc. | |
1609 | handle == X:Y - clear. | |
1610 | handle == X:0 - root class. | |
1611 | */ | |
1612 | ||
1613 | /* Step 1. Determine qdisc handle X:0 */ | |
1614 | ||
de179c8c H |
1615 | portid = tcm->tcm_parent; |
1616 | clid = tcm->tcm_handle; | |
1617 | qid = TC_H_MAJ(clid); | |
1618 | ||
15e47304 EB |
1619 | if (portid != TC_H_ROOT) { |
1620 | u32 qid1 = TC_H_MAJ(portid); | |
1da177e4 LT |
1621 | |
1622 | if (qid && qid1) { | |
1623 | /* If both majors are known, they must be identical. */ | |
1624 | if (qid != qid1) | |
1625 | return -EINVAL; | |
1626 | } else if (qid1) { | |
1627 | qid = qid1; | |
1628 | } else if (qid == 0) | |
af356afa | 1629 | qid = dev->qdisc->handle; |
1da177e4 LT |
1630 | |
1631 | /* Now qid is genuine qdisc handle consistent | |
cc7ec456 ED |
1632 | * both with parent and child. |
1633 | * | |
15e47304 | 1634 | * TC_H_MAJ(portid) still may be unspecified, complete it now. |
1da177e4 | 1635 | */ |
15e47304 EB |
1636 | if (portid) |
1637 | portid = TC_H_MAKE(qid, portid); | |
1da177e4 LT |
1638 | } else { |
1639 | if (qid == 0) | |
af356afa | 1640 | qid = dev->qdisc->handle; |
1da177e4 LT |
1641 | } |
1642 | ||
1643 | /* OK. Locate qdisc */ | |
cc7ec456 ED |
1644 | q = qdisc_lookup(dev, qid); |
1645 | if (!q) | |
1da177e4 LT |
1646 | return -ENOENT; |
1647 | ||
1648 | /* An check that it supports classes */ | |
1649 | cops = q->ops->cl_ops; | |
1650 | if (cops == NULL) | |
1651 | return -EINVAL; | |
1652 | ||
1653 | /* Now try to get class */ | |
1654 | if (clid == 0) { | |
15e47304 | 1655 | if (portid == TC_H_ROOT) |
1da177e4 LT |
1656 | clid = qid; |
1657 | } else | |
1658 | clid = TC_H_MAKE(qid, clid); | |
1659 | ||
1660 | if (clid) | |
1661 | cl = cops->get(q, clid); | |
1662 | ||
1663 | if (cl == 0) { | |
1664 | err = -ENOENT; | |
cc7ec456 ED |
1665 | if (n->nlmsg_type != RTM_NEWTCLASS || |
1666 | !(n->nlmsg_flags & NLM_F_CREATE)) | |
1da177e4 LT |
1667 | goto out; |
1668 | } else { | |
1669 | switch (n->nlmsg_type) { | |
10297b99 | 1670 | case RTM_NEWTCLASS: |
1da177e4 | 1671 | err = -EEXIST; |
cc7ec456 | 1672 | if (n->nlmsg_flags & NLM_F_EXCL) |
1da177e4 LT |
1673 | goto out; |
1674 | break; | |
1675 | case RTM_DELTCLASS: | |
de6d5cdf PM |
1676 | err = -EOPNOTSUPP; |
1677 | if (cops->delete) | |
1678 | err = cops->delete(q, cl); | |
1da177e4 | 1679 | if (err == 0) |
5a7a5555 JHS |
1680 | tclass_notify(net, skb, n, q, cl, |
1681 | RTM_DELTCLASS); | |
1da177e4 LT |
1682 | goto out; |
1683 | case RTM_GETTCLASS: | |
7316ae88 | 1684 | err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS); |
1da177e4 LT |
1685 | goto out; |
1686 | default: | |
1687 | err = -EINVAL; | |
1688 | goto out; | |
1689 | } | |
1690 | } | |
1691 | ||
1692 | new_cl = cl; | |
de6d5cdf PM |
1693 | err = -EOPNOTSUPP; |
1694 | if (cops->change) | |
15e47304 | 1695 | err = cops->change(q, clid, portid, tca, &new_cl); |
1da177e4 | 1696 | if (err == 0) |
7316ae88 | 1697 | tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS); |
1da177e4 LT |
1698 | |
1699 | out: | |
1700 | if (cl) | |
1701 | cops->put(q, cl); | |
1702 | ||
1703 | return err; | |
1704 | } | |
1705 | ||
1706 | ||
1707 | static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, | |
1708 | unsigned long cl, | |
15e47304 | 1709 | u32 portid, u32 seq, u16 flags, int event) |
1da177e4 LT |
1710 | { |
1711 | struct tcmsg *tcm; | |
1712 | struct nlmsghdr *nlh; | |
27a884dc | 1713 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 | 1714 | struct gnet_dump d; |
20fea08b | 1715 | const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; |
1da177e4 | 1716 | |
fba373d2 | 1717 | cond_resched(); |
15e47304 | 1718 | nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); |
02ef22ca DM |
1719 | if (!nlh) |
1720 | goto out_nlmsg_trim; | |
1721 | tcm = nlmsg_data(nlh); | |
1da177e4 | 1722 | tcm->tcm_family = AF_UNSPEC; |
16ebb5e0 ED |
1723 | tcm->tcm__pad1 = 0; |
1724 | tcm->tcm__pad2 = 0; | |
5ce2d488 | 1725 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex; |
1da177e4 LT |
1726 | tcm->tcm_parent = q->handle; |
1727 | tcm->tcm_handle = q->handle; | |
1728 | tcm->tcm_info = 0; | |
1b34ec43 DM |
1729 | if (nla_put_string(skb, TCA_KIND, q->ops->id)) |
1730 | goto nla_put_failure; | |
1da177e4 | 1731 | if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) |
1e90474c | 1732 | goto nla_put_failure; |
1da177e4 | 1733 | |
102396ae | 1734 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, |
edb09eb1 | 1735 | NULL, &d, TCA_PAD) < 0) |
1e90474c | 1736 | goto nla_put_failure; |
1da177e4 LT |
1737 | |
1738 | if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) | |
1e90474c | 1739 | goto nla_put_failure; |
1da177e4 LT |
1740 | |
1741 | if (gnet_stats_finish_copy(&d) < 0) | |
1e90474c | 1742 | goto nla_put_failure; |
1da177e4 | 1743 | |
27a884dc | 1744 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
1745 | return skb->len; |
1746 | ||
02ef22ca | 1747 | out_nlmsg_trim: |
1e90474c | 1748 | nla_put_failure: |
dc5fc579 | 1749 | nlmsg_trim(skb, b); |
1da177e4 LT |
1750 | return -1; |
1751 | } | |
1752 | ||
7316ae88 TG |
1753 | static int tclass_notify(struct net *net, struct sk_buff *oskb, |
1754 | struct nlmsghdr *n, struct Qdisc *q, | |
1755 | unsigned long cl, int event) | |
1da177e4 LT |
1756 | { |
1757 | struct sk_buff *skb; | |
15e47304 | 1758 | u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; |
1da177e4 LT |
1759 | |
1760 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1761 | if (!skb) | |
1762 | return -ENOBUFS; | |
1763 | ||
15e47304 | 1764 | if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) { |
1da177e4 LT |
1765 | kfree_skb(skb); |
1766 | return -EINVAL; | |
1767 | } | |
1768 | ||
15e47304 | 1769 | return rtnetlink_send(skb, net, portid, RTNLGRP_TC, |
cc7ec456 | 1770 | n->nlmsg_flags & NLM_F_ECHO); |
1da177e4 LT |
1771 | } |
1772 | ||
cc7ec456 ED |
1773 | struct qdisc_dump_args { |
1774 | struct qdisc_walker w; | |
1775 | struct sk_buff *skb; | |
1776 | struct netlink_callback *cb; | |
1da177e4 LT |
1777 | }; |
1778 | ||
5a7a5555 JHS |
1779 | static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, |
1780 | struct qdisc_walker *arg) | |
1da177e4 LT |
1781 | { |
1782 | struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; | |
1783 | ||
15e47304 | 1784 | return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid, |
5a7a5555 JHS |
1785 | a->cb->nlh->nlmsg_seq, NLM_F_MULTI, |
1786 | RTM_NEWTCLASS); | |
1da177e4 LT |
1787 | } |
1788 | ||
30723673 DM |
1789 | static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, |
1790 | struct tcmsg *tcm, struct netlink_callback *cb, | |
1791 | int *t_p, int s_t) | |
1792 | { | |
1793 | struct qdisc_dump_args arg; | |
1794 | ||
49b49971 | 1795 | if (tc_qdisc_dump_ignore(q, false) || |
30723673 DM |
1796 | *t_p < s_t || !q->ops->cl_ops || |
1797 | (tcm->tcm_parent && | |
1798 | TC_H_MAJ(tcm->tcm_parent) != q->handle)) { | |
1799 | (*t_p)++; | |
1800 | return 0; | |
1801 | } | |
1802 | if (*t_p > s_t) | |
1803 | memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); | |
1804 | arg.w.fn = qdisc_class_dump; | |
1805 | arg.skb = skb; | |
1806 | arg.cb = cb; | |
1807 | arg.w.stop = 0; | |
1808 | arg.w.skip = cb->args[1]; | |
1809 | arg.w.count = 0; | |
1810 | q->ops->cl_ops->walk(q, &arg.w); | |
1811 | cb->args[1] = arg.w.count; | |
1812 | if (arg.w.stop) | |
1813 | return -1; | |
1814 | (*t_p)++; | |
1815 | return 0; | |
1816 | } | |
1817 | ||
1818 | static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, | |
1819 | struct tcmsg *tcm, struct netlink_callback *cb, | |
1820 | int *t_p, int s_t) | |
1821 | { | |
1822 | struct Qdisc *q; | |
59cc1f61 | 1823 | int b; |
30723673 DM |
1824 | |
1825 | if (!root) | |
1826 | return 0; | |
1827 | ||
1828 | if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) | |
1829 | return -1; | |
1830 | ||
69012ae4 JK |
1831 | if (!qdisc_dev(root)) |
1832 | return 0; | |
1833 | ||
59cc1f61 | 1834 | hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { |
30723673 DM |
1835 | if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) |
1836 | return -1; | |
1837 | } | |
1838 | ||
1839 | return 0; | |
1840 | } | |
1841 | ||
1da177e4 LT |
1842 | static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) |
1843 | { | |
02ef22ca | 1844 | struct tcmsg *tcm = nlmsg_data(cb->nlh); |
3b1e0a65 | 1845 | struct net *net = sock_net(skb->sk); |
30723673 | 1846 | struct netdev_queue *dev_queue; |
1da177e4 | 1847 | struct net_device *dev; |
30723673 | 1848 | int t, s_t; |
1da177e4 | 1849 | |
573ce260 | 1850 | if (nlmsg_len(cb->nlh) < sizeof(*tcm)) |
1da177e4 | 1851 | return 0; |
cc7ec456 ED |
1852 | dev = dev_get_by_index(net, tcm->tcm_ifindex); |
1853 | if (!dev) | |
1da177e4 LT |
1854 | return 0; |
1855 | ||
1856 | s_t = cb->args[0]; | |
1857 | t = 0; | |
1858 | ||
af356afa | 1859 | if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) |
30723673 DM |
1860 | goto done; |
1861 | ||
24824a09 ED |
1862 | dev_queue = dev_ingress_queue(dev); |
1863 | if (dev_queue && | |
1864 | tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, | |
1865 | &t, s_t) < 0) | |
30723673 | 1866 | goto done; |
1da177e4 | 1867 | |
30723673 | 1868 | done: |
1da177e4 LT |
1869 | cb->args[0] = t; |
1870 | ||
1871 | dev_put(dev); | |
1872 | return skb->len; | |
1873 | } | |
1874 | ||
1875 | /* Main classifier routine: scans classifier chain attached | |
cc7ec456 ED |
1876 | * to this qdisc, (optionally) tests for protocol and asks |
1877 | * specific classifiers. | |
1da177e4 | 1878 | */ |
3b3ae880 DB |
1879 | int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp, |
1880 | struct tcf_result *res, bool compat_mode) | |
73ca4918 | 1881 | { |
d8b9605d | 1882 | __be16 protocol = tc_skb_protocol(skb); |
3b3ae880 | 1883 | #ifdef CONFIG_NET_CLS_ACT |
d6264071 | 1884 | const int max_reclassify_loop = 4; |
3b3ae880 DB |
1885 | const struct tcf_proto *old_tp = tp; |
1886 | int limit = 0; | |
73ca4918 | 1887 | |
3b3ae880 DB |
1888 | reclassify: |
1889 | #endif | |
25d8c0d5 | 1890 | for (; tp; tp = rcu_dereference_bh(tp->next)) { |
3b3ae880 DB |
1891 | int err; |
1892 | ||
cc7ec456 ED |
1893 | if (tp->protocol != protocol && |
1894 | tp->protocol != htons(ETH_P_ALL)) | |
1895 | continue; | |
cc7ec456 | 1896 | |
3b3ae880 DB |
1897 | err = tp->classify(skb, tp, res); |
1898 | #ifdef CONFIG_NET_CLS_ACT | |
c1b3b199 | 1899 | if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) |
3b3ae880 DB |
1900 | goto reset; |
1901 | #endif | |
e578d9c0 | 1902 | if (err >= 0) |
73ca4918 | 1903 | return err; |
73ca4918 | 1904 | } |
73ca4918 | 1905 | |
7e6e18fb | 1906 | return TC_ACT_UNSPEC; /* signal: continue lookup */ |
1da177e4 | 1907 | #ifdef CONFIG_NET_CLS_ACT |
3b3ae880 | 1908 | reset: |
d6264071 | 1909 | if (unlikely(limit++ >= max_reclassify_loop)) { |
c1b3b199 DB |
1910 | net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n", |
1911 | tp->q->ops->id, tp->prio & 0xffff, | |
1912 | ntohs(tp->protocol)); | |
3b3ae880 | 1913 | return TC_ACT_SHOT; |
1da177e4 | 1914 | } |
3b3ae880 DB |
1915 | |
1916 | tp = old_tp; | |
619fe326 | 1917 | protocol = tc_skb_protocol(skb); |
3b3ae880 | 1918 | goto reclassify; |
73ca4918 | 1919 | #endif |
1da177e4 | 1920 | } |
73ca4918 | 1921 | EXPORT_SYMBOL(tc_classify); |
1da177e4 | 1922 | |
1da177e4 LT |
1923 | #ifdef CONFIG_PROC_FS |
1924 | static int psched_show(struct seq_file *seq, void *v) | |
1925 | { | |
1926 | seq_printf(seq, "%08x %08x %08x %08x\n", | |
ca44d6e6 | 1927 | (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), |
514bca32 | 1928 | 1000000, |
1e317688 | 1929 | (u32)NSEC_PER_SEC / hrtimer_resolution); |
1da177e4 LT |
1930 | |
1931 | return 0; | |
1932 | } | |
1933 | ||
1934 | static int psched_open(struct inode *inode, struct file *file) | |
1935 | { | |
7e5ab157 | 1936 | return single_open(file, psched_show, NULL); |
1da177e4 LT |
1937 | } |
1938 | ||
da7071d7 | 1939 | static const struct file_operations psched_fops = { |
1da177e4 LT |
1940 | .owner = THIS_MODULE, |
1941 | .open = psched_open, | |
1942 | .read = seq_read, | |
1943 | .llseek = seq_lseek, | |
1944 | .release = single_release, | |
10297b99 | 1945 | }; |
7316ae88 TG |
1946 | |
1947 | static int __net_init psched_net_init(struct net *net) | |
1948 | { | |
1949 | struct proc_dir_entry *e; | |
1950 | ||
d4beaa66 | 1951 | e = proc_create("psched", 0, net->proc_net, &psched_fops); |
7316ae88 TG |
1952 | if (e == NULL) |
1953 | return -ENOMEM; | |
1954 | ||
1955 | return 0; | |
1956 | } | |
1957 | ||
1958 | static void __net_exit psched_net_exit(struct net *net) | |
1959 | { | |
ece31ffd | 1960 | remove_proc_entry("psched", net->proc_net); |
7316ae88 TG |
1961 | } |
1962 | #else | |
1963 | static int __net_init psched_net_init(struct net *net) | |
1964 | { | |
1965 | return 0; | |
1966 | } | |
1967 | ||
1968 | static void __net_exit psched_net_exit(struct net *net) | |
1969 | { | |
1970 | } | |
1da177e4 LT |
1971 | #endif |
1972 | ||
7316ae88 TG |
1973 | static struct pernet_operations psched_net_ops = { |
1974 | .init = psched_net_init, | |
1975 | .exit = psched_net_exit, | |
1976 | }; | |
1977 | ||
1da177e4 LT |
1978 | static int __init pktsched_init(void) |
1979 | { | |
7316ae88 TG |
1980 | int err; |
1981 | ||
1982 | err = register_pernet_subsys(&psched_net_ops); | |
1983 | if (err) { | |
cc7ec456 | 1984 | pr_err("pktsched_init: " |
7316ae88 TG |
1985 | "cannot initialize per netns operations\n"); |
1986 | return err; | |
1987 | } | |
1988 | ||
6da7c8fc | 1989 | register_qdisc(&pfifo_fast_ops); |
1da177e4 LT |
1990 | register_qdisc(&pfifo_qdisc_ops); |
1991 | register_qdisc(&bfifo_qdisc_ops); | |
57dbb2d8 | 1992 | register_qdisc(&pfifo_head_drop_qdisc_ops); |
6ec1c69a | 1993 | register_qdisc(&mq_qdisc_ops); |
d66d6c31 | 1994 | register_qdisc(&noqueue_qdisc_ops); |
1da177e4 | 1995 | |
c7ac8679 GR |
1996 | rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL); |
1997 | rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL); | |
5a7a5555 JHS |
1998 | rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, |
1999 | NULL); | |
c7ac8679 GR |
2000 | rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL); |
2001 | rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL); | |
5a7a5555 JHS |
2002 | rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, |
2003 | NULL); | |
be577ddc | 2004 | |
1da177e4 LT |
2005 | return 0; |
2006 | } | |
2007 | ||
2008 | subsys_initcall(pktsched_init); |