2 * net/sched/sch_api.c Packet scheduler API.
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.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
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
18 #include <linux/module.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/string.h>
22 #include <linux/errno.h>
23 #include <linux/skbuff.h>
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>
29 #include <linux/hrtimer.h>
30 #include <linux/lockdep.h>
31 #include <linux/slab.h>
33 #include <net/net_namespace.h>
35 #include <net/netlink.h>
36 #include <net/pkt_sched.h>
38 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
39 struct nlmsghdr *n, u32 clid,
40 struct Qdisc *old, struct Qdisc *new);
41 static int tclass_notify(struct net *net, struct sk_buff *oskb,
42 struct nlmsghdr *n, struct Qdisc *q,
43 unsigned long cl, int event);
50 This file consists of two interrelated parts:
52 1. queueing disciplines manager frontend.
53 2. traffic classes manager frontend.
55 Generally, queueing discipline ("qdisc") is a black box,
56 which is able to enqueue packets and to dequeue them (when
57 device is ready to send something) in order and at times
58 determined by algorithm hidden in it.
60 qdisc's are divided to two categories:
61 - "queues", which have no internal structure visible from outside.
62 - "schedulers", which split all the packets to "traffic classes",
63 using "packet classifiers" (look at cls_api.c)
65 In turn, classes may have child qdiscs (as rule, queues)
66 attached to them etc. etc. etc.
68 The goal of the routines in this file is to translate
69 information supplied by user in the form of handles
70 to more intelligible for kernel form, to make some sanity
71 checks and part of work, which is common to all qdiscs
72 and to provide rtnetlink notifications.
74 All real intelligent work is done inside qdisc modules.
78 Every discipline has two major routines: enqueue and dequeue.
82 dequeue usually returns a skb to send. It is allowed to return NULL,
83 but it does not mean that queue is empty, it just means that
84 discipline does not want to send anything this time.
85 Queue is really empty if q->q.qlen == 0.
86 For complicated disciplines with multiple queues q->q is not
87 real packet queue, but however q->q.qlen must be valid.
91 enqueue returns 0, if packet was enqueued successfully.
92 If packet (this one or another one) was dropped, it returns
94 NET_XMIT_DROP - this packet dropped
95 Expected action: do not backoff, but wait until queue will clear.
96 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
97 Expected action: backoff or ignore
98 NET_XMIT_POLICED - dropped by police.
99 Expected action: backoff or error to real-time apps.
105 like dequeue but without removing a packet from the queue
109 returns qdisc to initial state: purge all buffers, clear all
110 timers, counters (except for statistics) etc.
114 initializes newly created qdisc.
118 destroys resources allocated by init and during lifetime of qdisc.
122 changes qdisc parameters.
125 /* Protects list of registered TC modules. It is pure SMP lock. */
126 static DEFINE_RWLOCK(qdisc_mod_lock);
129 /************************************************
130 * Queueing disciplines manipulation. *
131 ************************************************/
134 /* The list of all installed queueing disciplines. */
136 static struct Qdisc_ops *qdisc_base;
138 /* Register/uregister queueing discipline */
140 int register_qdisc(struct Qdisc_ops *qops)
142 struct Qdisc_ops *q, **qp;
145 write_lock(&qdisc_mod_lock);
146 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
147 if (!strcmp(qops->id, q->id))
150 if (qops->enqueue == NULL)
151 qops->enqueue = noop_qdisc_ops.enqueue;
152 if (qops->peek == NULL) {
153 if (qops->dequeue == NULL)
154 qops->peek = noop_qdisc_ops.peek;
158 if (qops->dequeue == NULL)
159 qops->dequeue = noop_qdisc_ops.dequeue;
162 const struct Qdisc_class_ops *cops = qops->cl_ops;
164 if (!(cops->get && cops->put && cops->walk && cops->leaf))
167 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf))
175 write_unlock(&qdisc_mod_lock);
182 EXPORT_SYMBOL(register_qdisc);
184 int unregister_qdisc(struct Qdisc_ops *qops)
186 struct Qdisc_ops *q, **qp;
189 write_lock(&qdisc_mod_lock);
190 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
198 write_unlock(&qdisc_mod_lock);
201 EXPORT_SYMBOL(unregister_qdisc);
203 /* Get default qdisc if not otherwise specified */
204 void qdisc_get_default(char *name, size_t len)
206 read_lock(&qdisc_mod_lock);
207 strlcpy(name, default_qdisc_ops->id, len);
208 read_unlock(&qdisc_mod_lock);
211 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
213 struct Qdisc_ops *q = NULL;
215 for (q = qdisc_base; q; q = q->next) {
216 if (!strcmp(name, q->id)) {
217 if (!try_module_get(q->owner))
226 /* Set new default qdisc to use */
227 int qdisc_set_default(const char *name)
229 const struct Qdisc_ops *ops;
231 if (!capable(CAP_NET_ADMIN))
234 write_lock(&qdisc_mod_lock);
235 ops = qdisc_lookup_default(name);
237 /* Not found, drop lock and try to load module */
238 write_unlock(&qdisc_mod_lock);
239 request_module("sch_%s", name);
240 write_lock(&qdisc_mod_lock);
242 ops = qdisc_lookup_default(name);
246 /* Set new default */
247 module_put(default_qdisc_ops->owner);
248 default_qdisc_ops = ops;
250 write_unlock(&qdisc_mod_lock);
252 return ops ? 0 : -ENOENT;
255 /* We know handle. Find qdisc among all qdisc's attached to device
256 (root qdisc, all its children, children of children etc.)
259 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
263 if (!(root->flags & TCQ_F_BUILTIN) &&
264 root->handle == handle)
267 list_for_each_entry(q, &root->list, list) {
268 if (q->handle == handle)
274 static void qdisc_list_add(struct Qdisc *q)
276 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
277 list_add_tail(&q->list, &qdisc_dev(q)->qdisc->list);
280 void qdisc_list_del(struct Qdisc *q)
282 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
285 EXPORT_SYMBOL(qdisc_list_del);
287 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
291 q = qdisc_match_from_root(dev->qdisc, handle);
295 if (dev_ingress_queue(dev))
296 q = qdisc_match_from_root(
297 dev_ingress_queue(dev)->qdisc_sleeping,
303 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
307 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
311 cl = cops->get(p, classid);
315 leaf = cops->leaf(p, cl);
320 /* Find queueing discipline by name */
322 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
324 struct Qdisc_ops *q = NULL;
327 read_lock(&qdisc_mod_lock);
328 for (q = qdisc_base; q; q = q->next) {
329 if (nla_strcmp(kind, q->id) == 0) {
330 if (!try_module_get(q->owner))
335 read_unlock(&qdisc_mod_lock);
340 /* The linklayer setting were not transferred from iproute2, in older
341 * versions, and the rate tables lookup systems have been dropped in
342 * the kernel. To keep backward compatible with older iproute2 tc
343 * utils, we detect the linklayer setting by detecting if the rate
344 * table were modified.
346 * For linklayer ATM table entries, the rate table will be aligned to
347 * 48 bytes, thus some table entries will contain the same value. The
348 * mpu (min packet unit) is also encoded into the old rate table, thus
349 * starting from the mpu, we find low and high table entries for
350 * mapping this cell. If these entries contain the same value, when
351 * the rate tables have been modified for linklayer ATM.
353 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
354 * and then roundup to the next cell, calc the table entry one below,
357 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
359 int low = roundup(r->mpu, 48);
360 int high = roundup(low+1, 48);
361 int cell_low = low >> r->cell_log;
362 int cell_high = (high >> r->cell_log) - 1;
364 /* rtab is too inaccurate at rates > 100Mbit/s */
365 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
366 pr_debug("TC linklayer: Giving up ATM detection\n");
367 return TC_LINKLAYER_ETHERNET;
370 if ((cell_high > cell_low) && (cell_high < 256)
371 && (rtab[cell_low] == rtab[cell_high])) {
372 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
373 cell_low, cell_high, rtab[cell_high]);
374 return TC_LINKLAYER_ATM;
376 return TC_LINKLAYER_ETHERNET;
379 static struct qdisc_rate_table *qdisc_rtab_list;
381 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
383 struct qdisc_rate_table *rtab;
385 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
386 nla_len(tab) != TC_RTAB_SIZE)
389 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
390 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
391 !memcmp(&rtab->data, nla_data(tab), 1024)) {
397 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
401 memcpy(rtab->data, nla_data(tab), 1024);
402 if (r->linklayer == TC_LINKLAYER_UNAWARE)
403 r->linklayer = __detect_linklayer(r, rtab->data);
404 rtab->next = qdisc_rtab_list;
405 qdisc_rtab_list = rtab;
409 EXPORT_SYMBOL(qdisc_get_rtab);
411 void qdisc_put_rtab(struct qdisc_rate_table *tab)
413 struct qdisc_rate_table *rtab, **rtabp;
415 if (!tab || --tab->refcnt)
418 for (rtabp = &qdisc_rtab_list;
419 (rtab = *rtabp) != NULL;
420 rtabp = &rtab->next) {
428 EXPORT_SYMBOL(qdisc_put_rtab);
430 static LIST_HEAD(qdisc_stab_list);
431 static DEFINE_SPINLOCK(qdisc_stab_lock);
433 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
434 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
435 [TCA_STAB_DATA] = { .type = NLA_BINARY },
438 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
440 struct nlattr *tb[TCA_STAB_MAX + 1];
441 struct qdisc_size_table *stab;
442 struct tc_sizespec *s;
443 unsigned int tsize = 0;
447 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
450 if (!tb[TCA_STAB_BASE])
451 return ERR_PTR(-EINVAL);
453 s = nla_data(tb[TCA_STAB_BASE]);
456 if (!tb[TCA_STAB_DATA])
457 return ERR_PTR(-EINVAL);
458 tab = nla_data(tb[TCA_STAB_DATA]);
459 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
462 if (tsize != s->tsize || (!tab && tsize > 0))
463 return ERR_PTR(-EINVAL);
465 spin_lock(&qdisc_stab_lock);
467 list_for_each_entry(stab, &qdisc_stab_list, list) {
468 if (memcmp(&stab->szopts, s, sizeof(*s)))
470 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
473 spin_unlock(&qdisc_stab_lock);
477 spin_unlock(&qdisc_stab_lock);
479 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
481 return ERR_PTR(-ENOMEM);
486 memcpy(stab->data, tab, tsize * sizeof(u16));
488 spin_lock(&qdisc_stab_lock);
489 list_add_tail(&stab->list, &qdisc_stab_list);
490 spin_unlock(&qdisc_stab_lock);
495 static void stab_kfree_rcu(struct rcu_head *head)
497 kfree(container_of(head, struct qdisc_size_table, rcu));
500 void qdisc_put_stab(struct qdisc_size_table *tab)
505 spin_lock(&qdisc_stab_lock);
507 if (--tab->refcnt == 0) {
508 list_del(&tab->list);
509 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
512 spin_unlock(&qdisc_stab_lock);
514 EXPORT_SYMBOL(qdisc_put_stab);
516 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
520 nest = nla_nest_start(skb, TCA_STAB);
522 goto nla_put_failure;
523 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
524 goto nla_put_failure;
525 nla_nest_end(skb, nest);
533 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
537 pkt_len = skb->len + stab->szopts.overhead;
538 if (unlikely(!stab->szopts.tsize))
541 slot = pkt_len + stab->szopts.cell_align;
542 if (unlikely(slot < 0))
545 slot >>= stab->szopts.cell_log;
546 if (likely(slot < stab->szopts.tsize))
547 pkt_len = stab->data[slot];
549 pkt_len = stab->data[stab->szopts.tsize - 1] *
550 (slot / stab->szopts.tsize) +
551 stab->data[slot % stab->szopts.tsize];
553 pkt_len <<= stab->szopts.size_log;
555 if (unlikely(pkt_len < 1))
557 qdisc_skb_cb(skb)->pkt_len = pkt_len;
559 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
561 void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
563 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
564 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
565 txt, qdisc->ops->id, qdisc->handle >> 16);
566 qdisc->flags |= TCQ_F_WARN_NONWC;
569 EXPORT_SYMBOL(qdisc_warn_nonwc);
571 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
573 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
576 qdisc_unthrottled(wd->qdisc);
577 __netif_schedule(qdisc_root(wd->qdisc));
579 return HRTIMER_NORESTART;
582 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
584 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
585 wd->timer.function = qdisc_watchdog;
588 EXPORT_SYMBOL(qdisc_watchdog_init);
590 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
592 if (test_bit(__QDISC_STATE_DEACTIVATED,
593 &qdisc_root_sleeping(wd->qdisc)->state))
596 qdisc_throttled(wd->qdisc);
598 hrtimer_start(&wd->timer,
599 ns_to_ktime(expires),
602 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
604 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
606 hrtimer_cancel(&wd->timer);
607 qdisc_unthrottled(wd->qdisc);
609 EXPORT_SYMBOL(qdisc_watchdog_cancel);
611 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
613 unsigned int size = n * sizeof(struct hlist_head), i;
614 struct hlist_head *h;
616 if (size <= PAGE_SIZE)
617 h = kmalloc(size, GFP_KERNEL);
619 h = (struct hlist_head *)
620 __get_free_pages(GFP_KERNEL, get_order(size));
623 for (i = 0; i < n; i++)
624 INIT_HLIST_HEAD(&h[i]);
629 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
631 unsigned int size = n * sizeof(struct hlist_head);
633 if (size <= PAGE_SIZE)
636 free_pages((unsigned long)h, get_order(size));
639 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
641 struct Qdisc_class_common *cl;
642 struct hlist_node *next;
643 struct hlist_head *nhash, *ohash;
644 unsigned int nsize, nmask, osize;
647 /* Rehash when load factor exceeds 0.75 */
648 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
650 nsize = clhash->hashsize * 2;
652 nhash = qdisc_class_hash_alloc(nsize);
656 ohash = clhash->hash;
657 osize = clhash->hashsize;
660 for (i = 0; i < osize; i++) {
661 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
662 h = qdisc_class_hash(cl->classid, nmask);
663 hlist_add_head(&cl->hnode, &nhash[h]);
666 clhash->hash = nhash;
667 clhash->hashsize = nsize;
668 clhash->hashmask = nmask;
669 sch_tree_unlock(sch);
671 qdisc_class_hash_free(ohash, osize);
673 EXPORT_SYMBOL(qdisc_class_hash_grow);
675 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
677 unsigned int size = 4;
679 clhash->hash = qdisc_class_hash_alloc(size);
680 if (clhash->hash == NULL)
682 clhash->hashsize = size;
683 clhash->hashmask = size - 1;
684 clhash->hashelems = 0;
687 EXPORT_SYMBOL(qdisc_class_hash_init);
689 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
691 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
693 EXPORT_SYMBOL(qdisc_class_hash_destroy);
695 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
696 struct Qdisc_class_common *cl)
700 INIT_HLIST_NODE(&cl->hnode);
701 h = qdisc_class_hash(cl->classid, clhash->hashmask);
702 hlist_add_head(&cl->hnode, &clhash->hash[h]);
705 EXPORT_SYMBOL(qdisc_class_hash_insert);
707 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
708 struct Qdisc_class_common *cl)
710 hlist_del(&cl->hnode);
713 EXPORT_SYMBOL(qdisc_class_hash_remove);
715 /* Allocate an unique handle from space managed by kernel
716 * Possible range is [8000-FFFF]:0000 (0x8000 values)
718 static u32 qdisc_alloc_handle(struct net_device *dev)
721 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
724 autohandle += TC_H_MAKE(0x10000U, 0);
725 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
726 autohandle = TC_H_MAKE(0x80000000U, 0);
727 if (!qdisc_lookup(dev, autohandle))
735 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
737 const struct Qdisc_class_ops *cops;
744 drops = max_t(int, n, 0);
745 while ((parentid = sch->parent)) {
746 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
749 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
751 WARN_ON(parentid != TC_H_ROOT);
754 cops = sch->ops->cl_ops;
755 if (cops->qlen_notify) {
756 cl = cops->get(sch, parentid);
757 cops->qlen_notify(sch, cl);
761 sch->qstats.drops += drops;
764 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
766 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
767 struct nlmsghdr *n, u32 clid,
768 struct Qdisc *old, struct Qdisc *new)
771 qdisc_notify(net, skb, n, clid, old, new);
777 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
780 * When appropriate send a netlink notification using 'skb'
783 * On success, destroy old qdisc.
786 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
787 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
788 struct Qdisc *new, struct Qdisc *old)
790 struct Qdisc *q = old;
791 struct net *net = dev_net(dev);
794 if (parent == NULL) {
795 unsigned int i, num_q, ingress;
798 num_q = dev->num_tx_queues;
799 if ((q && q->flags & TCQ_F_INGRESS) ||
800 (new && new->flags & TCQ_F_INGRESS)) {
803 if (!dev_ingress_queue(dev))
807 if (dev->flags & IFF_UP)
810 if (new && new->ops->attach) {
811 new->ops->attach(new);
815 for (i = 0; i < num_q; i++) {
816 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
819 dev_queue = netdev_get_tx_queue(dev, i);
821 old = dev_graft_qdisc(dev_queue, new);
823 atomic_inc(&new->refcnt);
830 notify_and_destroy(net, skb, n, classid,
832 if (new && !new->ops->attach)
833 atomic_inc(&new->refcnt);
834 dev->qdisc = new ? : &noop_qdisc;
836 notify_and_destroy(net, skb, n, classid, old, new);
839 if (dev->flags & IFF_UP)
842 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
845 if (cops && cops->graft) {
846 unsigned long cl = cops->get(parent, classid);
848 err = cops->graft(parent, cl, new, &old);
849 cops->put(parent, cl);
854 notify_and_destroy(net, skb, n, classid, old, new);
859 /* lockdep annotation is needed for ingress; egress gets it only for name */
860 static struct lock_class_key qdisc_tx_lock;
861 static struct lock_class_key qdisc_rx_lock;
864 Allocate and initialize new qdisc.
866 Parameters are passed via opt.
869 static struct Qdisc *
870 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
871 struct Qdisc *p, u32 parent, u32 handle,
872 struct nlattr **tca, int *errp)
875 struct nlattr *kind = tca[TCA_KIND];
877 struct Qdisc_ops *ops;
878 struct qdisc_size_table *stab;
880 ops = qdisc_lookup_ops(kind);
881 #ifdef CONFIG_MODULES
882 if (ops == NULL && kind != NULL) {
884 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
885 /* We dropped the RTNL semaphore in order to
886 * perform the module load. So, even if we
887 * succeeded in loading the module we have to
888 * tell the caller to replay the request. We
889 * indicate this using -EAGAIN.
890 * We replay the request because the device may
891 * go away in the mean time.
894 request_module("sch_%s", name);
896 ops = qdisc_lookup_ops(kind);
898 /* We will try again qdisc_lookup_ops,
899 * so don't keep a reference.
901 module_put(ops->owner);
913 sch = qdisc_alloc(dev_queue, ops);
919 sch->parent = parent;
921 if (handle == TC_H_INGRESS) {
922 sch->flags |= TCQ_F_INGRESS;
923 handle = TC_H_MAKE(TC_H_INGRESS, 0);
924 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
927 handle = qdisc_alloc_handle(dev);
932 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
933 if (!netif_is_multiqueue(dev))
934 sch->flags |= TCQ_F_ONETXQUEUE;
937 sch->handle = handle;
939 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
941 stab = qdisc_get_stab(tca[TCA_STAB]);
946 rcu_assign_pointer(sch->stab, stab);
949 spinlock_t *root_lock;
952 if (sch->flags & TCQ_F_MQROOT)
955 if ((sch->parent != TC_H_ROOT) &&
956 !(sch->flags & TCQ_F_INGRESS) &&
957 (!p || !(p->flags & TCQ_F_MQROOT)))
958 root_lock = qdisc_root_sleeping_lock(sch);
960 root_lock = qdisc_lock(sch);
962 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
963 root_lock, tca[TCA_RATE]);
974 kfree((char *) sch - sch->padded);
976 module_put(ops->owner);
983 * Any broken qdiscs that would require a ops->reset() here?
984 * The qdisc was never in action so it shouldn't be necessary.
986 qdisc_put_stab(rtnl_dereference(sch->stab));
992 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
994 struct qdisc_size_table *ostab, *stab = NULL;
997 if (tca[TCA_OPTIONS]) {
998 if (sch->ops->change == NULL)
1000 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1005 if (tca[TCA_STAB]) {
1006 stab = qdisc_get_stab(tca[TCA_STAB]);
1008 return PTR_ERR(stab);
1011 ostab = rtnl_dereference(sch->stab);
1012 rcu_assign_pointer(sch->stab, stab);
1013 qdisc_put_stab(ostab);
1015 if (tca[TCA_RATE]) {
1016 /* NB: ignores errors from replace_estimator
1017 because change can't be undone. */
1018 if (sch->flags & TCQ_F_MQROOT)
1020 gen_replace_estimator(&sch->bstats, &sch->rate_est,
1021 qdisc_root_sleeping_lock(sch),
1028 struct check_loop_arg {
1029 struct qdisc_walker w;
1034 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
1036 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1038 struct check_loop_arg arg;
1040 if (q->ops->cl_ops == NULL)
1043 arg.w.stop = arg.w.skip = arg.w.count = 0;
1044 arg.w.fn = check_loop_fn;
1047 q->ops->cl_ops->walk(q, &arg.w);
1048 return arg.w.stop ? -ELOOP : 0;
1052 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1055 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1056 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1058 leaf = cops->leaf(q, cl);
1060 if (leaf == arg->p || arg->depth > 7)
1062 return check_loop(leaf, arg->p, arg->depth + 1);
1071 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1073 struct net *net = sock_net(skb->sk);
1074 struct tcmsg *tcm = nlmsg_data(n);
1075 struct nlattr *tca[TCA_MAX + 1];
1076 struct net_device *dev;
1078 struct Qdisc *q = NULL;
1079 struct Qdisc *p = NULL;
1082 if ((n->nlmsg_type != RTM_GETQDISC) && !capable(CAP_NET_ADMIN))
1085 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1089 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1093 clid = tcm->tcm_parent;
1095 if (clid != TC_H_ROOT) {
1096 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1097 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1100 q = qdisc_leaf(p, clid);
1101 } else if (dev_ingress_queue(dev)) {
1102 q = dev_ingress_queue(dev)->qdisc_sleeping;
1110 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1113 q = qdisc_lookup(dev, tcm->tcm_handle);
1118 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1121 if (n->nlmsg_type == RTM_DELQDISC) {
1126 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1130 qdisc_notify(net, skb, n, clid, NULL, q);
1136 * Create/change qdisc.
1139 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1141 struct net *net = sock_net(skb->sk);
1143 struct nlattr *tca[TCA_MAX + 1];
1144 struct net_device *dev;
1146 struct Qdisc *q, *p;
1149 if (!capable(CAP_NET_ADMIN))
1153 /* Reinit, just in case something touches this. */
1154 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1158 tcm = nlmsg_data(n);
1159 clid = tcm->tcm_parent;
1162 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1168 if (clid != TC_H_ROOT) {
1169 if (clid != TC_H_INGRESS) {
1170 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1173 q = qdisc_leaf(p, clid);
1174 } else if (dev_ingress_queue_create(dev)) {
1175 q = dev_ingress_queue(dev)->qdisc_sleeping;
1181 /* It may be default qdisc, ignore it */
1182 if (q && q->handle == 0)
1185 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1186 if (tcm->tcm_handle) {
1187 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1189 if (TC_H_MIN(tcm->tcm_handle))
1191 q = qdisc_lookup(dev, tcm->tcm_handle);
1193 goto create_n_graft;
1194 if (n->nlmsg_flags & NLM_F_EXCL)
1196 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1199 (p && check_loop(q, p, 0)))
1201 atomic_inc(&q->refcnt);
1205 goto create_n_graft;
1207 /* This magic test requires explanation.
1209 * We know, that some child q is already
1210 * attached to this parent and have choice:
1211 * either to change it or to create/graft new one.
1213 * 1. We are allowed to create/graft only
1214 * if CREATE and REPLACE flags are set.
1216 * 2. If EXCL is set, requestor wanted to say,
1217 * that qdisc tcm_handle is not expected
1218 * to exist, so that we choose create/graft too.
1220 * 3. The last case is when no flags are set.
1221 * Alas, it is sort of hole in API, we
1222 * cannot decide what to do unambiguously.
1223 * For now we select create/graft, if
1224 * user gave KIND, which does not match existing.
1226 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1227 (n->nlmsg_flags & NLM_F_REPLACE) &&
1228 ((n->nlmsg_flags & NLM_F_EXCL) ||
1230 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1231 goto create_n_graft;
1235 if (!tcm->tcm_handle)
1237 q = qdisc_lookup(dev, tcm->tcm_handle);
1240 /* Change qdisc parameters */
1243 if (n->nlmsg_flags & NLM_F_EXCL)
1245 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1247 err = qdisc_change(q, tca);
1249 qdisc_notify(net, skb, n, clid, NULL, q);
1253 if (!(n->nlmsg_flags & NLM_F_CREATE))
1255 if (clid == TC_H_INGRESS) {
1256 if (dev_ingress_queue(dev))
1257 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1258 tcm->tcm_parent, tcm->tcm_parent,
1263 struct netdev_queue *dev_queue;
1265 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1266 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1268 dev_queue = p->dev_queue;
1270 dev_queue = netdev_get_tx_queue(dev, 0);
1272 q = qdisc_create(dev, dev_queue, p,
1273 tcm->tcm_parent, tcm->tcm_handle,
1283 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1293 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1294 u32 portid, u32 seq, u16 flags, int event)
1297 struct nlmsghdr *nlh;
1298 unsigned char *b = skb_tail_pointer(skb);
1300 struct qdisc_size_table *stab;
1302 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1304 goto out_nlmsg_trim;
1305 tcm = nlmsg_data(nlh);
1306 tcm->tcm_family = AF_UNSPEC;
1309 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1310 tcm->tcm_parent = clid;
1311 tcm->tcm_handle = q->handle;
1312 tcm->tcm_info = atomic_read(&q->refcnt);
1313 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1314 goto nla_put_failure;
1315 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1316 goto nla_put_failure;
1317 q->qstats.qlen = q->q.qlen;
1319 stab = rtnl_dereference(q->stab);
1320 if (stab && qdisc_dump_stab(skb, stab) < 0)
1321 goto nla_put_failure;
1323 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1324 qdisc_root_sleeping_lock(q), &d) < 0)
1325 goto nla_put_failure;
1327 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1328 goto nla_put_failure;
1330 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1331 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1332 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1333 goto nla_put_failure;
1335 if (gnet_stats_finish_copy(&d) < 0)
1336 goto nla_put_failure;
1338 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1347 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1349 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1352 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1353 struct nlmsghdr *n, u32 clid,
1354 struct Qdisc *old, struct Qdisc *new)
1356 struct sk_buff *skb;
1357 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1359 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1363 if (old && !tc_qdisc_dump_ignore(old)) {
1364 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1365 0, RTM_DELQDISC) < 0)
1368 if (new && !tc_qdisc_dump_ignore(new)) {
1369 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1370 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1375 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1376 n->nlmsg_flags & NLM_F_ECHO);
1383 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1384 struct netlink_callback *cb,
1385 int *q_idx_p, int s_q_idx)
1387 int ret = 0, q_idx = *q_idx_p;
1394 if (q_idx < s_q_idx) {
1397 if (!tc_qdisc_dump_ignore(q) &&
1398 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1399 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1403 list_for_each_entry(q, &root->list, list) {
1404 if (q_idx < s_q_idx) {
1408 if (!tc_qdisc_dump_ignore(q) &&
1409 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1410 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1423 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1425 struct net *net = sock_net(skb->sk);
1428 struct net_device *dev;
1430 s_idx = cb->args[0];
1431 s_q_idx = q_idx = cb->args[1];
1435 for_each_netdev_rcu(net, dev) {
1436 struct netdev_queue *dev_queue;
1444 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1447 dev_queue = dev_ingress_queue(dev);
1449 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1450 &q_idx, s_q_idx) < 0)
1461 cb->args[1] = q_idx;
1468 /************************************************
1469 * Traffic classes manipulation. *
1470 ************************************************/
1474 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1476 struct net *net = sock_net(skb->sk);
1477 struct tcmsg *tcm = nlmsg_data(n);
1478 struct nlattr *tca[TCA_MAX + 1];
1479 struct net_device *dev;
1480 struct Qdisc *q = NULL;
1481 const struct Qdisc_class_ops *cops;
1482 unsigned long cl = 0;
1483 unsigned long new_cl;
1489 if ((n->nlmsg_type != RTM_GETTCLASS) && !capable(CAP_NET_ADMIN))
1492 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1496 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1501 parent == TC_H_UNSPEC - unspecified parent.
1502 parent == TC_H_ROOT - class is root, which has no parent.
1503 parent == X:0 - parent is root class.
1504 parent == X:Y - parent is a node in hierarchy.
1505 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1507 handle == 0:0 - generate handle from kernel pool.
1508 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1509 handle == X:Y - clear.
1510 handle == X:0 - root class.
1513 /* Step 1. Determine qdisc handle X:0 */
1515 portid = tcm->tcm_parent;
1516 clid = tcm->tcm_handle;
1517 qid = TC_H_MAJ(clid);
1519 if (portid != TC_H_ROOT) {
1520 u32 qid1 = TC_H_MAJ(portid);
1523 /* If both majors are known, they must be identical. */
1528 } else if (qid == 0)
1529 qid = dev->qdisc->handle;
1531 /* Now qid is genuine qdisc handle consistent
1532 * both with parent and child.
1534 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1537 portid = TC_H_MAKE(qid, portid);
1540 qid = dev->qdisc->handle;
1543 /* OK. Locate qdisc */
1544 q = qdisc_lookup(dev, qid);
1548 /* An check that it supports classes */
1549 cops = q->ops->cl_ops;
1553 /* Now try to get class */
1555 if (portid == TC_H_ROOT)
1558 clid = TC_H_MAKE(qid, clid);
1561 cl = cops->get(q, clid);
1565 if (n->nlmsg_type != RTM_NEWTCLASS ||
1566 !(n->nlmsg_flags & NLM_F_CREATE))
1569 switch (n->nlmsg_type) {
1572 if (n->nlmsg_flags & NLM_F_EXCL)
1578 err = cops->delete(q, cl);
1580 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1583 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1594 err = cops->change(q, clid, portid, tca, &new_cl);
1596 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1606 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1608 u32 portid, u32 seq, u16 flags, int event)
1611 struct nlmsghdr *nlh;
1612 unsigned char *b = skb_tail_pointer(skb);
1614 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1616 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1618 goto out_nlmsg_trim;
1619 tcm = nlmsg_data(nlh);
1620 tcm->tcm_family = AF_UNSPEC;
1623 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1624 tcm->tcm_parent = q->handle;
1625 tcm->tcm_handle = q->handle;
1627 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1628 goto nla_put_failure;
1629 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1630 goto nla_put_failure;
1632 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1633 qdisc_root_sleeping_lock(q), &d) < 0)
1634 goto nla_put_failure;
1636 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1637 goto nla_put_failure;
1639 if (gnet_stats_finish_copy(&d) < 0)
1640 goto nla_put_failure;
1642 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1651 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1652 struct nlmsghdr *n, struct Qdisc *q,
1653 unsigned long cl, int event)
1655 struct sk_buff *skb;
1656 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1658 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1662 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1667 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1668 n->nlmsg_flags & NLM_F_ECHO);
1671 struct qdisc_dump_args {
1672 struct qdisc_walker w;
1673 struct sk_buff *skb;
1674 struct netlink_callback *cb;
1677 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1679 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1681 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1682 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1685 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1686 struct tcmsg *tcm, struct netlink_callback *cb,
1689 struct qdisc_dump_args arg;
1691 if (tc_qdisc_dump_ignore(q) ||
1692 *t_p < s_t || !q->ops->cl_ops ||
1694 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1699 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1700 arg.w.fn = qdisc_class_dump;
1704 arg.w.skip = cb->args[1];
1706 q->ops->cl_ops->walk(q, &arg.w);
1707 cb->args[1] = arg.w.count;
1714 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1715 struct tcmsg *tcm, struct netlink_callback *cb,
1723 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1726 list_for_each_entry(q, &root->list, list) {
1727 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1734 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1736 struct tcmsg *tcm = nlmsg_data(cb->nlh);
1737 struct net *net = sock_net(skb->sk);
1738 struct netdev_queue *dev_queue;
1739 struct net_device *dev;
1742 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1744 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1751 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1754 dev_queue = dev_ingress_queue(dev);
1756 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1767 /* Main classifier routine: scans classifier chain attached
1768 * to this qdisc, (optionally) tests for protocol and asks
1769 * specific classifiers.
1771 int tc_classify_compat(struct sk_buff *skb, const struct tcf_proto *tp,
1772 struct tcf_result *res)
1774 __be16 protocol = skb->protocol;
1777 for (; tp; tp = tp->next) {
1778 if (tp->protocol != protocol &&
1779 tp->protocol != htons(ETH_P_ALL))
1781 err = tp->classify(skb, tp, res);
1784 #ifdef CONFIG_NET_CLS_ACT
1785 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1786 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1793 EXPORT_SYMBOL(tc_classify_compat);
1795 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1796 struct tcf_result *res)
1799 #ifdef CONFIG_NET_CLS_ACT
1800 const struct tcf_proto *otp = tp;
1804 err = tc_classify_compat(skb, tp, res);
1805 #ifdef CONFIG_NET_CLS_ACT
1806 if (err == TC_ACT_RECLASSIFY) {
1807 u32 verd = G_TC_VERD(skb->tc_verd);
1810 if (verd++ >= MAX_REC_LOOP) {
1811 net_notice_ratelimited("%s: packet reclassify loop rule prio %u protocol %02x\n",
1814 ntohs(tp->protocol));
1817 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1823 EXPORT_SYMBOL(tc_classify);
1825 void tcf_destroy(struct tcf_proto *tp)
1827 tp->ops->destroy(tp);
1828 module_put(tp->ops->owner);
1832 void tcf_destroy_chain(struct tcf_proto **fl)
1834 struct tcf_proto *tp;
1836 while ((tp = *fl) != NULL) {
1841 EXPORT_SYMBOL(tcf_destroy_chain);
1843 #ifdef CONFIG_PROC_FS
1844 static int psched_show(struct seq_file *seq, void *v)
1848 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1849 seq_printf(seq, "%08x %08x %08x %08x\n",
1850 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1852 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1857 static int psched_open(struct inode *inode, struct file *file)
1859 return single_open(file, psched_show, NULL);
1862 static const struct file_operations psched_fops = {
1863 .owner = THIS_MODULE,
1864 .open = psched_open,
1866 .llseek = seq_lseek,
1867 .release = single_release,
1870 static int __net_init psched_net_init(struct net *net)
1872 struct proc_dir_entry *e;
1874 e = proc_create("psched", 0, net->proc_net, &psched_fops);
1881 static void __net_exit psched_net_exit(struct net *net)
1883 remove_proc_entry("psched", net->proc_net);
1886 static int __net_init psched_net_init(struct net *net)
1891 static void __net_exit psched_net_exit(struct net *net)
1896 static struct pernet_operations psched_net_ops = {
1897 .init = psched_net_init,
1898 .exit = psched_net_exit,
1901 static int __init pktsched_init(void)
1905 err = register_pernet_subsys(&psched_net_ops);
1907 pr_err("pktsched_init: "
1908 "cannot initialize per netns operations\n");
1912 register_qdisc(&pfifo_fast_ops);
1913 register_qdisc(&pfifo_qdisc_ops);
1914 register_qdisc(&bfifo_qdisc_ops);
1915 register_qdisc(&pfifo_head_drop_qdisc_ops);
1916 register_qdisc(&mq_qdisc_ops);
1918 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1919 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1920 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1921 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1922 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1923 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1928 subsys_initcall(pktsched_init);