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/unregister 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.)
257 * Note: caller either uses rtnl or rcu_read_lock()
260 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
264 if (!(root->flags & TCQ_F_BUILTIN) &&
265 root->handle == handle)
268 list_for_each_entry_rcu(q, &root->list, list) {
269 if (q->handle == handle)
275 void qdisc_list_add(struct Qdisc *q)
277 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
278 struct Qdisc *root = qdisc_dev(q)->qdisc;
280 WARN_ON_ONCE(root == &noop_qdisc);
282 list_add_tail_rcu(&q->list, &root->list);
285 EXPORT_SYMBOL(qdisc_list_add);
287 void qdisc_list_del(struct Qdisc *q)
289 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
291 list_del_rcu(&q->list);
294 EXPORT_SYMBOL(qdisc_list_del);
296 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
300 q = qdisc_match_from_root(dev->qdisc, handle);
304 if (dev_ingress_queue(dev))
305 q = qdisc_match_from_root(
306 dev_ingress_queue(dev)->qdisc_sleeping,
312 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
316 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
320 cl = cops->get(p, classid);
324 leaf = cops->leaf(p, cl);
329 /* Find queueing discipline by name */
331 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
333 struct Qdisc_ops *q = NULL;
336 read_lock(&qdisc_mod_lock);
337 for (q = qdisc_base; q; q = q->next) {
338 if (nla_strcmp(kind, q->id) == 0) {
339 if (!try_module_get(q->owner))
344 read_unlock(&qdisc_mod_lock);
349 /* The linklayer setting were not transferred from iproute2, in older
350 * versions, and the rate tables lookup systems have been dropped in
351 * the kernel. To keep backward compatible with older iproute2 tc
352 * utils, we detect the linklayer setting by detecting if the rate
353 * table were modified.
355 * For linklayer ATM table entries, the rate table will be aligned to
356 * 48 bytes, thus some table entries will contain the same value. The
357 * mpu (min packet unit) is also encoded into the old rate table, thus
358 * starting from the mpu, we find low and high table entries for
359 * mapping this cell. If these entries contain the same value, when
360 * the rate tables have been modified for linklayer ATM.
362 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
363 * and then roundup to the next cell, calc the table entry one below,
366 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
368 int low = roundup(r->mpu, 48);
369 int high = roundup(low+1, 48);
370 int cell_low = low >> r->cell_log;
371 int cell_high = (high >> r->cell_log) - 1;
373 /* rtab is too inaccurate at rates > 100Mbit/s */
374 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
375 pr_debug("TC linklayer: Giving up ATM detection\n");
376 return TC_LINKLAYER_ETHERNET;
379 if ((cell_high > cell_low) && (cell_high < 256)
380 && (rtab[cell_low] == rtab[cell_high])) {
381 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
382 cell_low, cell_high, rtab[cell_high]);
383 return TC_LINKLAYER_ATM;
385 return TC_LINKLAYER_ETHERNET;
388 static struct qdisc_rate_table *qdisc_rtab_list;
390 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
392 struct qdisc_rate_table *rtab;
394 if (tab == NULL || r->rate == 0 ||
395 r->cell_log == 0 || r->cell_log >= 32 ||
396 nla_len(tab) != TC_RTAB_SIZE)
399 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
400 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
401 !memcmp(&rtab->data, nla_data(tab), 1024)) {
407 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
411 memcpy(rtab->data, nla_data(tab), 1024);
412 if (r->linklayer == TC_LINKLAYER_UNAWARE)
413 r->linklayer = __detect_linklayer(r, rtab->data);
414 rtab->next = qdisc_rtab_list;
415 qdisc_rtab_list = rtab;
419 EXPORT_SYMBOL(qdisc_get_rtab);
421 void qdisc_put_rtab(struct qdisc_rate_table *tab)
423 struct qdisc_rate_table *rtab, **rtabp;
425 if (!tab || --tab->refcnt)
428 for (rtabp = &qdisc_rtab_list;
429 (rtab = *rtabp) != NULL;
430 rtabp = &rtab->next) {
438 EXPORT_SYMBOL(qdisc_put_rtab);
440 static LIST_HEAD(qdisc_stab_list);
441 static DEFINE_SPINLOCK(qdisc_stab_lock);
443 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
444 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
445 [TCA_STAB_DATA] = { .type = NLA_BINARY },
448 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
450 struct nlattr *tb[TCA_STAB_MAX + 1];
451 struct qdisc_size_table *stab;
452 struct tc_sizespec *s;
453 unsigned int tsize = 0;
457 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
460 if (!tb[TCA_STAB_BASE])
461 return ERR_PTR(-EINVAL);
463 s = nla_data(tb[TCA_STAB_BASE]);
466 if (!tb[TCA_STAB_DATA])
467 return ERR_PTR(-EINVAL);
468 tab = nla_data(tb[TCA_STAB_DATA]);
469 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
472 if (tsize != s->tsize || (!tab && tsize > 0))
473 return ERR_PTR(-EINVAL);
475 spin_lock(&qdisc_stab_lock);
477 list_for_each_entry(stab, &qdisc_stab_list, list) {
478 if (memcmp(&stab->szopts, s, sizeof(*s)))
480 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
483 spin_unlock(&qdisc_stab_lock);
487 spin_unlock(&qdisc_stab_lock);
489 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
491 return ERR_PTR(-ENOMEM);
496 memcpy(stab->data, tab, tsize * sizeof(u16));
498 spin_lock(&qdisc_stab_lock);
499 list_add_tail(&stab->list, &qdisc_stab_list);
500 spin_unlock(&qdisc_stab_lock);
505 static void stab_kfree_rcu(struct rcu_head *head)
507 kfree(container_of(head, struct qdisc_size_table, rcu));
510 void qdisc_put_stab(struct qdisc_size_table *tab)
515 spin_lock(&qdisc_stab_lock);
517 if (--tab->refcnt == 0) {
518 list_del(&tab->list);
519 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
522 spin_unlock(&qdisc_stab_lock);
524 EXPORT_SYMBOL(qdisc_put_stab);
526 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
530 nest = nla_nest_start(skb, TCA_STAB);
532 goto nla_put_failure;
533 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
534 goto nla_put_failure;
535 nla_nest_end(skb, nest);
543 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
547 pkt_len = skb->len + stab->szopts.overhead;
548 if (unlikely(!stab->szopts.tsize))
551 slot = pkt_len + stab->szopts.cell_align;
552 if (unlikely(slot < 0))
555 slot >>= stab->szopts.cell_log;
556 if (likely(slot < stab->szopts.tsize))
557 pkt_len = stab->data[slot];
559 pkt_len = stab->data[stab->szopts.tsize - 1] *
560 (slot / stab->szopts.tsize) +
561 stab->data[slot % stab->szopts.tsize];
563 pkt_len <<= stab->szopts.size_log;
565 if (unlikely(pkt_len < 1))
567 qdisc_skb_cb(skb)->pkt_len = pkt_len;
569 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
571 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
573 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
574 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
575 txt, qdisc->ops->id, qdisc->handle >> 16);
576 qdisc->flags |= TCQ_F_WARN_NONWC;
579 EXPORT_SYMBOL(qdisc_warn_nonwc);
581 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
583 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
587 qdisc_unthrottled(wd->qdisc);
588 __netif_schedule(qdisc_root(wd->qdisc));
591 return HRTIMER_NORESTART;
594 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
596 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
597 wd->timer.function = qdisc_watchdog;
600 EXPORT_SYMBOL(qdisc_watchdog_init);
602 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires, bool throttle)
604 if (test_bit(__QDISC_STATE_DEACTIVATED,
605 &qdisc_root_sleeping(wd->qdisc)->state))
609 qdisc_throttled(wd->qdisc);
611 hrtimer_start(&wd->timer,
612 ns_to_ktime(expires),
613 HRTIMER_MODE_ABS_PINNED);
615 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
617 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
619 hrtimer_cancel(&wd->timer);
620 qdisc_unthrottled(wd->qdisc);
622 EXPORT_SYMBOL(qdisc_watchdog_cancel);
624 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
626 unsigned int size = n * sizeof(struct hlist_head), i;
627 struct hlist_head *h;
629 if (size <= PAGE_SIZE)
630 h = kmalloc(size, GFP_KERNEL);
632 h = (struct hlist_head *)
633 __get_free_pages(GFP_KERNEL, get_order(size));
636 for (i = 0; i < n; i++)
637 INIT_HLIST_HEAD(&h[i]);
642 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
644 unsigned int size = n * sizeof(struct hlist_head);
646 if (size <= PAGE_SIZE)
649 free_pages((unsigned long)h, get_order(size));
652 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
654 struct Qdisc_class_common *cl;
655 struct hlist_node *next;
656 struct hlist_head *nhash, *ohash;
657 unsigned int nsize, nmask, osize;
660 /* Rehash when load factor exceeds 0.75 */
661 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
663 nsize = clhash->hashsize * 2;
665 nhash = qdisc_class_hash_alloc(nsize);
669 ohash = clhash->hash;
670 osize = clhash->hashsize;
673 for (i = 0; i < osize; i++) {
674 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
675 h = qdisc_class_hash(cl->classid, nmask);
676 hlist_add_head(&cl->hnode, &nhash[h]);
679 clhash->hash = nhash;
680 clhash->hashsize = nsize;
681 clhash->hashmask = nmask;
682 sch_tree_unlock(sch);
684 qdisc_class_hash_free(ohash, osize);
686 EXPORT_SYMBOL(qdisc_class_hash_grow);
688 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
690 unsigned int size = 4;
692 clhash->hash = qdisc_class_hash_alloc(size);
693 if (clhash->hash == NULL)
695 clhash->hashsize = size;
696 clhash->hashmask = size - 1;
697 clhash->hashelems = 0;
700 EXPORT_SYMBOL(qdisc_class_hash_init);
702 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
704 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
706 EXPORT_SYMBOL(qdisc_class_hash_destroy);
708 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
709 struct Qdisc_class_common *cl)
713 INIT_HLIST_NODE(&cl->hnode);
714 h = qdisc_class_hash(cl->classid, clhash->hashmask);
715 hlist_add_head(&cl->hnode, &clhash->hash[h]);
718 EXPORT_SYMBOL(qdisc_class_hash_insert);
720 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
721 struct Qdisc_class_common *cl)
723 hlist_del(&cl->hnode);
726 EXPORT_SYMBOL(qdisc_class_hash_remove);
728 /* Allocate an unique handle from space managed by kernel
729 * Possible range is [8000-FFFF]:0000 (0x8000 values)
731 static u32 qdisc_alloc_handle(struct net_device *dev)
734 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
737 autohandle += TC_H_MAKE(0x10000U, 0);
738 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
739 autohandle = TC_H_MAKE(0x80000000U, 0);
740 if (!qdisc_lookup(dev, autohandle))
748 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n,
751 const struct Qdisc_class_ops *cops;
756 if (n == 0 && len == 0)
758 drops = max_t(int, n, 0);
760 while ((parentid = sch->parent)) {
761 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
764 if (sch->flags & TCQ_F_NOPARENT)
766 /* TODO: perform the search on a per txq basis */
767 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
769 WARN_ON_ONCE(parentid != TC_H_ROOT);
772 cops = sch->ops->cl_ops;
773 if (cops->qlen_notify) {
774 cl = cops->get(sch, parentid);
775 cops->qlen_notify(sch, cl);
779 sch->qstats.backlog -= len;
780 __qdisc_qstats_drop(sch, drops);
784 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
786 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
787 struct nlmsghdr *n, u32 clid,
788 struct Qdisc *old, struct Qdisc *new)
791 qdisc_notify(net, skb, n, clid, old, new);
797 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
800 * When appropriate send a netlink notification using 'skb'
803 * On success, destroy old qdisc.
806 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
807 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
808 struct Qdisc *new, struct Qdisc *old)
810 struct Qdisc *q = old;
811 struct net *net = dev_net(dev);
814 if (parent == NULL) {
815 unsigned int i, num_q, ingress;
818 num_q = dev->num_tx_queues;
819 if ((q && q->flags & TCQ_F_INGRESS) ||
820 (new && new->flags & TCQ_F_INGRESS)) {
823 if (!dev_ingress_queue(dev))
827 if (dev->flags & IFF_UP)
830 if (new && new->ops->attach)
833 for (i = 0; i < num_q; i++) {
834 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
837 dev_queue = netdev_get_tx_queue(dev, i);
839 old = dev_graft_qdisc(dev_queue, new);
841 atomic_inc(&new->refcnt);
849 notify_and_destroy(net, skb, n, classid,
851 if (new && !new->ops->attach)
852 atomic_inc(&new->refcnt);
853 dev->qdisc = new ? : &noop_qdisc;
855 if (new && new->ops->attach)
856 new->ops->attach(new);
858 notify_and_destroy(net, skb, n, classid, old, new);
861 if (dev->flags & IFF_UP)
864 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
867 if (cops && cops->graft) {
868 unsigned long cl = cops->get(parent, classid);
870 err = cops->graft(parent, cl, new, &old);
871 cops->put(parent, cl);
876 notify_and_destroy(net, skb, n, classid, old, new);
881 /* lockdep annotation is needed for ingress; egress gets it only for name */
882 static struct lock_class_key qdisc_tx_lock;
883 static struct lock_class_key qdisc_rx_lock;
886 Allocate and initialize new qdisc.
888 Parameters are passed via opt.
891 static struct Qdisc *
892 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
893 struct Qdisc *p, u32 parent, u32 handle,
894 struct nlattr **tca, int *errp)
897 struct nlattr *kind = tca[TCA_KIND];
899 struct Qdisc_ops *ops;
900 struct qdisc_size_table *stab;
902 ops = qdisc_lookup_ops(kind);
903 #ifdef CONFIG_MODULES
904 if (ops == NULL && kind != NULL) {
906 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
907 /* We dropped the RTNL semaphore in order to
908 * perform the module load. So, even if we
909 * succeeded in loading the module we have to
910 * tell the caller to replay the request. We
911 * indicate this using -EAGAIN.
912 * We replay the request because the device may
913 * go away in the mean time.
916 request_module("sch_%s", name);
918 ops = qdisc_lookup_ops(kind);
920 /* We will try again qdisc_lookup_ops,
921 * so don't keep a reference.
923 module_put(ops->owner);
935 sch = qdisc_alloc(dev_queue, ops);
941 sch->parent = parent;
943 if (handle == TC_H_INGRESS) {
944 sch->flags |= TCQ_F_INGRESS;
945 handle = TC_H_MAKE(TC_H_INGRESS, 0);
946 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
949 handle = qdisc_alloc_handle(dev);
954 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
955 if (!netif_is_multiqueue(dev))
956 sch->flags |= TCQ_F_ONETXQUEUE;
959 sch->handle = handle;
961 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
962 if (qdisc_is_percpu_stats(sch)) {
964 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
965 if (!sch->cpu_bstats)
968 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
969 if (!sch->cpu_qstats)
974 stab = qdisc_get_stab(tca[TCA_STAB]);
979 rcu_assign_pointer(sch->stab, stab);
982 spinlock_t *root_lock;
985 if (sch->flags & TCQ_F_MQROOT)
988 if ((sch->parent != TC_H_ROOT) &&
989 !(sch->flags & TCQ_F_INGRESS) &&
990 (!p || !(p->flags & TCQ_F_MQROOT)))
991 root_lock = qdisc_root_sleeping_lock(sch);
993 root_lock = qdisc_lock(sch);
995 err = gen_new_estimator(&sch->bstats,
1004 qdisc_list_add(sch);
1008 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1013 kfree((char *) sch - sch->padded);
1015 module_put(ops->owner);
1021 free_percpu(sch->cpu_bstats);
1022 free_percpu(sch->cpu_qstats);
1024 * Any broken qdiscs that would require a ops->reset() here?
1025 * The qdisc was never in action so it shouldn't be necessary.
1027 qdisc_put_stab(rtnl_dereference(sch->stab));
1033 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
1035 struct qdisc_size_table *ostab, *stab = NULL;
1038 if (tca[TCA_OPTIONS]) {
1039 if (sch->ops->change == NULL)
1041 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1046 if (tca[TCA_STAB]) {
1047 stab = qdisc_get_stab(tca[TCA_STAB]);
1049 return PTR_ERR(stab);
1052 ostab = rtnl_dereference(sch->stab);
1053 rcu_assign_pointer(sch->stab, stab);
1054 qdisc_put_stab(ostab);
1056 if (tca[TCA_RATE]) {
1057 /* NB: ignores errors from replace_estimator
1058 because change can't be undone. */
1059 if (sch->flags & TCQ_F_MQROOT)
1061 gen_replace_estimator(&sch->bstats,
1064 qdisc_root_sleeping_lock(sch),
1071 struct check_loop_arg {
1072 struct qdisc_walker w;
1077 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
1079 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1081 struct check_loop_arg arg;
1083 if (q->ops->cl_ops == NULL)
1086 arg.w.stop = arg.w.skip = arg.w.count = 0;
1087 arg.w.fn = check_loop_fn;
1090 q->ops->cl_ops->walk(q, &arg.w);
1091 return arg.w.stop ? -ELOOP : 0;
1095 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1098 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1099 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1101 leaf = cops->leaf(q, cl);
1103 if (leaf == arg->p || arg->depth > 7)
1105 return check_loop(leaf, arg->p, arg->depth + 1);
1114 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1116 struct net *net = sock_net(skb->sk);
1117 struct tcmsg *tcm = nlmsg_data(n);
1118 struct nlattr *tca[TCA_MAX + 1];
1119 struct net_device *dev;
1121 struct Qdisc *q = NULL;
1122 struct Qdisc *p = NULL;
1125 if ((n->nlmsg_type != RTM_GETQDISC) &&
1126 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1129 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1133 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1137 clid = tcm->tcm_parent;
1139 if (clid != TC_H_ROOT) {
1140 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1141 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1144 q = qdisc_leaf(p, clid);
1145 } else if (dev_ingress_queue(dev)) {
1146 q = dev_ingress_queue(dev)->qdisc_sleeping;
1154 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1157 q = qdisc_lookup(dev, tcm->tcm_handle);
1162 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1165 if (n->nlmsg_type == RTM_DELQDISC) {
1170 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1174 qdisc_notify(net, skb, n, clid, NULL, q);
1180 * Create/change qdisc.
1183 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1185 struct net *net = sock_net(skb->sk);
1187 struct nlattr *tca[TCA_MAX + 1];
1188 struct net_device *dev;
1190 struct Qdisc *q, *p;
1193 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1197 /* Reinit, just in case something touches this. */
1198 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1202 tcm = nlmsg_data(n);
1203 clid = tcm->tcm_parent;
1206 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1212 if (clid != TC_H_ROOT) {
1213 if (clid != TC_H_INGRESS) {
1214 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1217 q = qdisc_leaf(p, clid);
1218 } else if (dev_ingress_queue_create(dev)) {
1219 q = dev_ingress_queue(dev)->qdisc_sleeping;
1225 /* It may be default qdisc, ignore it */
1226 if (q && q->handle == 0)
1229 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1230 if (tcm->tcm_handle) {
1231 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1233 if (TC_H_MIN(tcm->tcm_handle))
1235 q = qdisc_lookup(dev, tcm->tcm_handle);
1237 goto create_n_graft;
1238 if (n->nlmsg_flags & NLM_F_EXCL)
1240 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1243 (p && check_loop(q, p, 0)))
1245 atomic_inc(&q->refcnt);
1249 goto create_n_graft;
1251 /* This magic test requires explanation.
1253 * We know, that some child q is already
1254 * attached to this parent and have choice:
1255 * either to change it or to create/graft new one.
1257 * 1. We are allowed to create/graft only
1258 * if CREATE and REPLACE flags are set.
1260 * 2. If EXCL is set, requestor wanted to say,
1261 * that qdisc tcm_handle is not expected
1262 * to exist, so that we choose create/graft too.
1264 * 3. The last case is when no flags are set.
1265 * Alas, it is sort of hole in API, we
1266 * cannot decide what to do unambiguously.
1267 * For now we select create/graft, if
1268 * user gave KIND, which does not match existing.
1270 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1271 (n->nlmsg_flags & NLM_F_REPLACE) &&
1272 ((n->nlmsg_flags & NLM_F_EXCL) ||
1274 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1275 goto create_n_graft;
1279 if (!tcm->tcm_handle)
1281 q = qdisc_lookup(dev, tcm->tcm_handle);
1284 /* Change qdisc parameters */
1287 if (n->nlmsg_flags & NLM_F_EXCL)
1289 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1291 err = qdisc_change(q, tca);
1293 qdisc_notify(net, skb, n, clid, NULL, q);
1297 if (!(n->nlmsg_flags & NLM_F_CREATE))
1299 if (clid == TC_H_INGRESS) {
1300 if (dev_ingress_queue(dev))
1301 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1302 tcm->tcm_parent, tcm->tcm_parent,
1307 struct netdev_queue *dev_queue;
1309 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1310 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1312 dev_queue = p->dev_queue;
1314 dev_queue = netdev_get_tx_queue(dev, 0);
1316 q = qdisc_create(dev, dev_queue, p,
1317 tcm->tcm_parent, tcm->tcm_handle,
1327 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1337 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1338 u32 portid, u32 seq, u16 flags, int event)
1340 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
1341 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
1343 struct nlmsghdr *nlh;
1344 unsigned char *b = skb_tail_pointer(skb);
1346 struct qdisc_size_table *stab;
1350 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1352 goto out_nlmsg_trim;
1353 tcm = nlmsg_data(nlh);
1354 tcm->tcm_family = AF_UNSPEC;
1357 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1358 tcm->tcm_parent = clid;
1359 tcm->tcm_handle = q->handle;
1360 tcm->tcm_info = atomic_read(&q->refcnt);
1361 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1362 goto nla_put_failure;
1363 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1364 goto nla_put_failure;
1367 stab = rtnl_dereference(q->stab);
1368 if (stab && qdisc_dump_stab(skb, stab) < 0)
1369 goto nla_put_failure;
1371 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1372 qdisc_root_sleeping_lock(q), &d) < 0)
1373 goto nla_put_failure;
1375 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1376 goto nla_put_failure;
1378 if (qdisc_is_percpu_stats(q)) {
1379 cpu_bstats = q->cpu_bstats;
1380 cpu_qstats = q->cpu_qstats;
1383 if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats) < 0 ||
1384 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1385 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
1386 goto nla_put_failure;
1388 if (gnet_stats_finish_copy(&d) < 0)
1389 goto nla_put_failure;
1391 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1400 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1402 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1405 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1406 struct nlmsghdr *n, u32 clid,
1407 struct Qdisc *old, struct Qdisc *new)
1409 struct sk_buff *skb;
1410 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1412 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1416 if (old && !tc_qdisc_dump_ignore(old)) {
1417 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1418 0, RTM_DELQDISC) < 0)
1421 if (new && !tc_qdisc_dump_ignore(new)) {
1422 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1423 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1428 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1429 n->nlmsg_flags & NLM_F_ECHO);
1436 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1437 struct netlink_callback *cb,
1438 int *q_idx_p, int s_q_idx)
1440 int ret = 0, q_idx = *q_idx_p;
1447 if (q_idx < s_q_idx) {
1450 if (!tc_qdisc_dump_ignore(q) &&
1451 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1452 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1456 list_for_each_entry(q, &root->list, list) {
1457 if (q_idx < s_q_idx) {
1461 if (!tc_qdisc_dump_ignore(q) &&
1462 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1463 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1476 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1478 struct net *net = sock_net(skb->sk);
1481 struct net_device *dev;
1483 s_idx = cb->args[0];
1484 s_q_idx = q_idx = cb->args[1];
1488 for_each_netdev(net, dev) {
1489 struct netdev_queue *dev_queue;
1497 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1500 dev_queue = dev_ingress_queue(dev);
1502 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1503 &q_idx, s_q_idx) < 0)
1512 cb->args[1] = q_idx;
1519 /************************************************
1520 * Traffic classes manipulation. *
1521 ************************************************/
1525 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1527 struct net *net = sock_net(skb->sk);
1528 struct tcmsg *tcm = nlmsg_data(n);
1529 struct nlattr *tca[TCA_MAX + 1];
1530 struct net_device *dev;
1531 struct Qdisc *q = NULL;
1532 const struct Qdisc_class_ops *cops;
1533 unsigned long cl = 0;
1534 unsigned long new_cl;
1540 if ((n->nlmsg_type != RTM_GETTCLASS) &&
1541 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1544 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1548 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1553 parent == TC_H_UNSPEC - unspecified parent.
1554 parent == TC_H_ROOT - class is root, which has no parent.
1555 parent == X:0 - parent is root class.
1556 parent == X:Y - parent is a node in hierarchy.
1557 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1559 handle == 0:0 - generate handle from kernel pool.
1560 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1561 handle == X:Y - clear.
1562 handle == X:0 - root class.
1565 /* Step 1. Determine qdisc handle X:0 */
1567 portid = tcm->tcm_parent;
1568 clid = tcm->tcm_handle;
1569 qid = TC_H_MAJ(clid);
1571 if (portid != TC_H_ROOT) {
1572 u32 qid1 = TC_H_MAJ(portid);
1575 /* If both majors are known, they must be identical. */
1580 } else if (qid == 0)
1581 qid = dev->qdisc->handle;
1583 /* Now qid is genuine qdisc handle consistent
1584 * both with parent and child.
1586 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1589 portid = TC_H_MAKE(qid, portid);
1592 qid = dev->qdisc->handle;
1595 /* OK. Locate qdisc */
1596 q = qdisc_lookup(dev, qid);
1600 /* An check that it supports classes */
1601 cops = q->ops->cl_ops;
1605 /* Now try to get class */
1607 if (portid == TC_H_ROOT)
1610 clid = TC_H_MAKE(qid, clid);
1613 cl = cops->get(q, clid);
1617 if (n->nlmsg_type != RTM_NEWTCLASS ||
1618 !(n->nlmsg_flags & NLM_F_CREATE))
1621 switch (n->nlmsg_type) {
1624 if (n->nlmsg_flags & NLM_F_EXCL)
1630 err = cops->delete(q, cl);
1632 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1635 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1646 err = cops->change(q, clid, portid, tca, &new_cl);
1648 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1658 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1660 u32 portid, u32 seq, u16 flags, int event)
1663 struct nlmsghdr *nlh;
1664 unsigned char *b = skb_tail_pointer(skb);
1666 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1669 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1671 goto out_nlmsg_trim;
1672 tcm = nlmsg_data(nlh);
1673 tcm->tcm_family = AF_UNSPEC;
1676 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1677 tcm->tcm_parent = q->handle;
1678 tcm->tcm_handle = q->handle;
1680 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1681 goto nla_put_failure;
1682 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1683 goto nla_put_failure;
1685 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1686 qdisc_root_sleeping_lock(q), &d) < 0)
1687 goto nla_put_failure;
1689 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1690 goto nla_put_failure;
1692 if (gnet_stats_finish_copy(&d) < 0)
1693 goto nla_put_failure;
1695 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1704 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1705 struct nlmsghdr *n, struct Qdisc *q,
1706 unsigned long cl, int event)
1708 struct sk_buff *skb;
1709 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1711 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1715 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1720 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1721 n->nlmsg_flags & NLM_F_ECHO);
1724 struct qdisc_dump_args {
1725 struct qdisc_walker w;
1726 struct sk_buff *skb;
1727 struct netlink_callback *cb;
1730 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1732 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1734 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1735 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1738 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1739 struct tcmsg *tcm, struct netlink_callback *cb,
1742 struct qdisc_dump_args arg;
1744 if (tc_qdisc_dump_ignore(q) ||
1745 *t_p < s_t || !q->ops->cl_ops ||
1747 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1752 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1753 arg.w.fn = qdisc_class_dump;
1757 arg.w.skip = cb->args[1];
1759 q->ops->cl_ops->walk(q, &arg.w);
1760 cb->args[1] = arg.w.count;
1767 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1768 struct tcmsg *tcm, struct netlink_callback *cb,
1776 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1779 list_for_each_entry(q, &root->list, list) {
1780 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1787 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1789 struct tcmsg *tcm = nlmsg_data(cb->nlh);
1790 struct net *net = sock_net(skb->sk);
1791 struct netdev_queue *dev_queue;
1792 struct net_device *dev;
1795 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1797 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1804 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1807 dev_queue = dev_ingress_queue(dev);
1809 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1820 /* Main classifier routine: scans classifier chain attached
1821 * to this qdisc, (optionally) tests for protocol and asks
1822 * specific classifiers.
1824 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1825 struct tcf_result *res, bool compat_mode)
1827 #ifdef CONFIG_NET_CLS_ACT
1828 const struct tcf_proto *old_tp = tp;
1833 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1834 __be16 protocol = tc_skb_protocol(skb);
1837 if (tp->protocol != protocol &&
1838 tp->protocol != htons(ETH_P_ALL))
1841 err = tp->classify(skb, tp, res);
1842 #ifdef CONFIG_NET_CLS_ACT
1843 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode))
1851 #ifdef CONFIG_NET_CLS_ACT
1853 if (unlikely(limit++ >= MAX_REC_LOOP)) {
1854 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
1855 tp->q->ops->id, tp->prio & 0xffff,
1856 ntohs(tp->protocol));
1864 EXPORT_SYMBOL(tc_classify);
1866 bool tcf_destroy(struct tcf_proto *tp, bool force)
1868 if (tp->ops->destroy(tp, force)) {
1869 module_put(tp->ops->owner);
1877 void tcf_destroy_chain(struct tcf_proto __rcu **fl)
1879 struct tcf_proto *tp;
1881 while ((tp = rtnl_dereference(*fl)) != NULL) {
1882 RCU_INIT_POINTER(*fl, tp->next);
1883 tcf_destroy(tp, true);
1886 EXPORT_SYMBOL(tcf_destroy_chain);
1888 #ifdef CONFIG_PROC_FS
1889 static int psched_show(struct seq_file *seq, void *v)
1891 seq_printf(seq, "%08x %08x %08x %08x\n",
1892 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1894 (u32)NSEC_PER_SEC / hrtimer_resolution);
1899 static int psched_open(struct inode *inode, struct file *file)
1901 return single_open(file, psched_show, NULL);
1904 static const struct file_operations psched_fops = {
1905 .owner = THIS_MODULE,
1906 .open = psched_open,
1908 .llseek = seq_lseek,
1909 .release = single_release,
1912 static int __net_init psched_net_init(struct net *net)
1914 struct proc_dir_entry *e;
1916 e = proc_create("psched", 0, net->proc_net, &psched_fops);
1923 static void __net_exit psched_net_exit(struct net *net)
1925 remove_proc_entry("psched", net->proc_net);
1928 static int __net_init psched_net_init(struct net *net)
1933 static void __net_exit psched_net_exit(struct net *net)
1938 static struct pernet_operations psched_net_ops = {
1939 .init = psched_net_init,
1940 .exit = psched_net_exit,
1943 static int __init pktsched_init(void)
1947 err = register_pernet_subsys(&psched_net_ops);
1949 pr_err("pktsched_init: "
1950 "cannot initialize per netns operations\n");
1954 register_qdisc(&pfifo_fast_ops);
1955 register_qdisc(&pfifo_qdisc_ops);
1956 register_qdisc(&bfifo_qdisc_ops);
1957 register_qdisc(&pfifo_head_drop_qdisc_ops);
1958 register_qdisc(&mq_qdisc_ops);
1959 register_qdisc(&noqueue_qdisc_ops);
1961 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1962 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1963 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1964 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1965 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1966 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1971 subsys_initcall(pktsched_init);