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>
32 #include <linux/hashtable.h>
34 #include <net/net_namespace.h>
36 #include <net/netlink.h>
37 #include <net/pkt_sched.h>
39 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
40 struct nlmsghdr *n, u32 clid,
41 struct Qdisc *old, struct Qdisc *new);
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);
51 This file consists of two interrelated parts:
53 1. queueing disciplines manager frontend.
54 2. traffic classes manager frontend.
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.
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)
66 In turn, classes may have child qdiscs (as rule, queues)
67 attached to them etc. etc. etc.
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.
75 All real intelligent work is done inside qdisc modules.
79 Every discipline has two major routines: enqueue and dequeue.
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.
92 enqueue returns 0, if packet was enqueued successfully.
93 If packet (this one or another one) was dropped, it returns
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
104 like dequeue but without removing a packet from the queue
108 returns qdisc to initial state: purge all buffers, clear all
109 timers, counters (except for statistics) etc.
113 initializes newly created qdisc.
117 destroys resources allocated by init and during lifetime of qdisc.
121 changes qdisc parameters.
124 /* Protects list of registered TC modules. It is pure SMP lock. */
125 static DEFINE_RWLOCK(qdisc_mod_lock);
128 /************************************************
129 * Queueing disciplines manipulation. *
130 ************************************************/
133 /* The list of all installed queueing disciplines. */
135 static struct Qdisc_ops *qdisc_base;
137 /* Register/unregister queueing discipline */
139 int register_qdisc(struct Qdisc_ops *qops)
141 struct Qdisc_ops *q, **qp;
144 write_lock(&qdisc_mod_lock);
145 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
146 if (!strcmp(qops->id, q->id))
149 if (qops->enqueue == NULL)
150 qops->enqueue = noop_qdisc_ops.enqueue;
151 if (qops->peek == NULL) {
152 if (qops->dequeue == NULL)
153 qops->peek = noop_qdisc_ops.peek;
157 if (qops->dequeue == NULL)
158 qops->dequeue = noop_qdisc_ops.dequeue;
161 const struct Qdisc_class_ops *cops = qops->cl_ops;
163 if (!(cops->get && cops->put && cops->walk && cops->leaf))
166 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf))
174 write_unlock(&qdisc_mod_lock);
181 EXPORT_SYMBOL(register_qdisc);
183 int unregister_qdisc(struct Qdisc_ops *qops)
185 struct Qdisc_ops *q, **qp;
188 write_lock(&qdisc_mod_lock);
189 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
197 write_unlock(&qdisc_mod_lock);
200 EXPORT_SYMBOL(unregister_qdisc);
202 /* Get default qdisc if not otherwise specified */
203 void qdisc_get_default(char *name, size_t len)
205 read_lock(&qdisc_mod_lock);
206 strlcpy(name, default_qdisc_ops->id, len);
207 read_unlock(&qdisc_mod_lock);
210 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
212 struct Qdisc_ops *q = NULL;
214 for (q = qdisc_base; q; q = q->next) {
215 if (!strcmp(name, q->id)) {
216 if (!try_module_get(q->owner))
225 /* Set new default qdisc to use */
226 int qdisc_set_default(const char *name)
228 const struct Qdisc_ops *ops;
230 if (!capable(CAP_NET_ADMIN))
233 write_lock(&qdisc_mod_lock);
234 ops = qdisc_lookup_default(name);
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);
241 ops = qdisc_lookup_default(name);
245 /* Set new default */
246 module_put(default_qdisc_ops->owner);
247 default_qdisc_ops = ops;
249 write_unlock(&qdisc_mod_lock);
251 return ops ? 0 : -ENOENT;
254 /* We know handle. Find qdisc among all qdisc's attached to device
255 * (root qdisc, all its children, children of children etc.)
256 * Note: caller either uses rtnl or rcu_read_lock()
259 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
263 if (!qdisc_dev(root))
264 return (root->handle == handle ? root : NULL);
266 if (!(root->flags & TCQ_F_BUILTIN) &&
267 root->handle == handle)
270 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) {
271 if (q->handle == handle)
277 void qdisc_hash_add(struct Qdisc *q)
279 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
281 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
284 EXPORT_SYMBOL(qdisc_hash_add);
286 void qdisc_hash_del(struct Qdisc *q)
288 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
290 hash_del_rcu(&q->hash);
293 EXPORT_SYMBOL(qdisc_hash_del);
295 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
299 q = qdisc_match_from_root(dev->qdisc, handle);
303 if (dev_ingress_queue(dev))
304 q = qdisc_match_from_root(
305 dev_ingress_queue(dev)->qdisc_sleeping,
311 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
315 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
319 cl = cops->get(p, classid);
323 leaf = cops->leaf(p, cl);
328 /* Find queueing discipline by name */
330 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
332 struct Qdisc_ops *q = NULL;
335 read_lock(&qdisc_mod_lock);
336 for (q = qdisc_base; q; q = q->next) {
337 if (nla_strcmp(kind, q->id) == 0) {
338 if (!try_module_get(q->owner))
343 read_unlock(&qdisc_mod_lock);
348 /* The linklayer setting were not transferred from iproute2, in older
349 * versions, and the rate tables lookup systems have been dropped in
350 * the kernel. To keep backward compatible with older iproute2 tc
351 * utils, we detect the linklayer setting by detecting if the rate
352 * table were modified.
354 * For linklayer ATM table entries, the rate table will be aligned to
355 * 48 bytes, thus some table entries will contain the same value. The
356 * mpu (min packet unit) is also encoded into the old rate table, thus
357 * starting from the mpu, we find low and high table entries for
358 * mapping this cell. If these entries contain the same value, when
359 * the rate tables have been modified for linklayer ATM.
361 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
362 * and then roundup to the next cell, calc the table entry one below,
365 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
367 int low = roundup(r->mpu, 48);
368 int high = roundup(low+1, 48);
369 int cell_low = low >> r->cell_log;
370 int cell_high = (high >> r->cell_log) - 1;
372 /* rtab is too inaccurate at rates > 100Mbit/s */
373 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
374 pr_debug("TC linklayer: Giving up ATM detection\n");
375 return TC_LINKLAYER_ETHERNET;
378 if ((cell_high > cell_low) && (cell_high < 256)
379 && (rtab[cell_low] == rtab[cell_high])) {
380 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
381 cell_low, cell_high, rtab[cell_high]);
382 return TC_LINKLAYER_ATM;
384 return TC_LINKLAYER_ETHERNET;
387 static struct qdisc_rate_table *qdisc_rtab_list;
389 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
392 struct qdisc_rate_table *rtab;
394 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
395 nla_len(tab) != TC_RTAB_SIZE)
398 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
399 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
400 !memcmp(&rtab->data, nla_data(tab), 1024)) {
406 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
410 memcpy(rtab->data, nla_data(tab), 1024);
411 if (r->linklayer == TC_LINKLAYER_UNAWARE)
412 r->linklayer = __detect_linklayer(r, rtab->data);
413 rtab->next = qdisc_rtab_list;
414 qdisc_rtab_list = rtab;
418 EXPORT_SYMBOL(qdisc_get_rtab);
420 void qdisc_put_rtab(struct qdisc_rate_table *tab)
422 struct qdisc_rate_table *rtab, **rtabp;
424 if (!tab || --tab->refcnt)
427 for (rtabp = &qdisc_rtab_list;
428 (rtab = *rtabp) != NULL;
429 rtabp = &rtab->next) {
437 EXPORT_SYMBOL(qdisc_put_rtab);
439 static LIST_HEAD(qdisc_stab_list);
440 static DEFINE_SPINLOCK(qdisc_stab_lock);
442 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
443 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
444 [TCA_STAB_DATA] = { .type = NLA_BINARY },
447 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
449 struct nlattr *tb[TCA_STAB_MAX + 1];
450 struct qdisc_size_table *stab;
451 struct tc_sizespec *s;
452 unsigned int tsize = 0;
456 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
459 if (!tb[TCA_STAB_BASE])
460 return ERR_PTR(-EINVAL);
462 s = nla_data(tb[TCA_STAB_BASE]);
465 if (!tb[TCA_STAB_DATA])
466 return ERR_PTR(-EINVAL);
467 tab = nla_data(tb[TCA_STAB_DATA]);
468 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
471 if (tsize != s->tsize || (!tab && tsize > 0))
472 return ERR_PTR(-EINVAL);
474 spin_lock(&qdisc_stab_lock);
476 list_for_each_entry(stab, &qdisc_stab_list, list) {
477 if (memcmp(&stab->szopts, s, sizeof(*s)))
479 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
482 spin_unlock(&qdisc_stab_lock);
486 spin_unlock(&qdisc_stab_lock);
488 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
490 return ERR_PTR(-ENOMEM);
495 memcpy(stab->data, tab, tsize * sizeof(u16));
497 spin_lock(&qdisc_stab_lock);
498 list_add_tail(&stab->list, &qdisc_stab_list);
499 spin_unlock(&qdisc_stab_lock);
504 static void stab_kfree_rcu(struct rcu_head *head)
506 kfree(container_of(head, struct qdisc_size_table, rcu));
509 void qdisc_put_stab(struct qdisc_size_table *tab)
514 spin_lock(&qdisc_stab_lock);
516 if (--tab->refcnt == 0) {
517 list_del(&tab->list);
518 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
521 spin_unlock(&qdisc_stab_lock);
523 EXPORT_SYMBOL(qdisc_put_stab);
525 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
529 nest = nla_nest_start(skb, TCA_STAB);
531 goto nla_put_failure;
532 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
533 goto nla_put_failure;
534 nla_nest_end(skb, nest);
542 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
543 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 __netif_schedule(qdisc_root(wd->qdisc));
590 return HRTIMER_NORESTART;
593 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
595 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
596 wd->timer.function = qdisc_watchdog;
599 EXPORT_SYMBOL(qdisc_watchdog_init);
601 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
603 if (test_bit(__QDISC_STATE_DEACTIVATED,
604 &qdisc_root_sleeping(wd->qdisc)->state))
607 if (wd->last_expires == expires)
610 wd->last_expires = expires;
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);
621 EXPORT_SYMBOL(qdisc_watchdog_cancel);
623 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
625 unsigned int size = n * sizeof(struct hlist_head), i;
626 struct hlist_head *h;
628 if (size <= PAGE_SIZE)
629 h = kmalloc(size, GFP_KERNEL);
631 h = (struct hlist_head *)
632 __get_free_pages(GFP_KERNEL, get_order(size));
635 for (i = 0; i < n; i++)
636 INIT_HLIST_HEAD(&h[i]);
641 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
643 unsigned int size = n * sizeof(struct hlist_head);
645 if (size <= PAGE_SIZE)
648 free_pages((unsigned long)h, get_order(size));
651 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
653 struct Qdisc_class_common *cl;
654 struct hlist_node *next;
655 struct hlist_head *nhash, *ohash;
656 unsigned int nsize, nmask, osize;
659 /* Rehash when load factor exceeds 0.75 */
660 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
662 nsize = clhash->hashsize * 2;
664 nhash = qdisc_class_hash_alloc(nsize);
668 ohash = clhash->hash;
669 osize = clhash->hashsize;
672 for (i = 0; i < osize; i++) {
673 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
674 h = qdisc_class_hash(cl->classid, nmask);
675 hlist_add_head(&cl->hnode, &nhash[h]);
678 clhash->hash = nhash;
679 clhash->hashsize = nsize;
680 clhash->hashmask = nmask;
681 sch_tree_unlock(sch);
683 qdisc_class_hash_free(ohash, osize);
685 EXPORT_SYMBOL(qdisc_class_hash_grow);
687 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
689 unsigned int size = 4;
691 clhash->hash = qdisc_class_hash_alloc(size);
692 if (clhash->hash == NULL)
694 clhash->hashsize = size;
695 clhash->hashmask = size - 1;
696 clhash->hashelems = 0;
699 EXPORT_SYMBOL(qdisc_class_hash_init);
701 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
703 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
705 EXPORT_SYMBOL(qdisc_class_hash_destroy);
707 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
708 struct Qdisc_class_common *cl)
712 INIT_HLIST_NODE(&cl->hnode);
713 h = qdisc_class_hash(cl->classid, clhash->hashmask);
714 hlist_add_head(&cl->hnode, &clhash->hash[h]);
717 EXPORT_SYMBOL(qdisc_class_hash_insert);
719 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
720 struct Qdisc_class_common *cl)
722 hlist_del(&cl->hnode);
725 EXPORT_SYMBOL(qdisc_class_hash_remove);
727 /* Allocate an unique handle from space managed by kernel
728 * Possible range is [8000-FFFF]:0000 (0x8000 values)
730 static u32 qdisc_alloc_handle(struct net_device *dev)
733 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
736 autohandle += TC_H_MAKE(0x10000U, 0);
737 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
738 autohandle = TC_H_MAKE(0x80000000U, 0);
739 if (!qdisc_lookup(dev, autohandle))
747 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n,
750 const struct Qdisc_class_ops *cops;
755 if (n == 0 && len == 0)
757 drops = max_t(int, n, 0);
759 while ((parentid = sch->parent)) {
760 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
763 if (sch->flags & TCQ_F_NOPARENT)
765 /* TODO: perform the search on a per txq basis */
766 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
768 WARN_ON_ONCE(parentid != TC_H_ROOT);
771 cops = sch->ops->cl_ops;
772 if (cops->qlen_notify) {
773 cl = cops->get(sch, parentid);
774 cops->qlen_notify(sch, cl);
778 sch->qstats.backlog -= len;
779 __qdisc_qstats_drop(sch, drops);
783 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
785 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
786 struct nlmsghdr *n, u32 clid,
787 struct Qdisc *old, struct Qdisc *new)
790 qdisc_notify(net, skb, n, clid, old, new);
796 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
799 * When appropriate send a netlink notification using 'skb'
802 * On success, destroy old qdisc.
805 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
806 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
807 struct Qdisc *new, struct Qdisc *old)
809 struct Qdisc *q = old;
810 struct net *net = dev_net(dev);
813 if (parent == NULL) {
814 unsigned int i, num_q, ingress;
817 num_q = dev->num_tx_queues;
818 if ((q && q->flags & TCQ_F_INGRESS) ||
819 (new && new->flags & TCQ_F_INGRESS)) {
822 if (!dev_ingress_queue(dev))
826 if (dev->flags & IFF_UP)
829 if (new && new->ops->attach)
832 for (i = 0; i < num_q; i++) {
833 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
836 dev_queue = netdev_get_tx_queue(dev, i);
838 old = dev_graft_qdisc(dev_queue, new);
840 atomic_inc(&new->refcnt);
848 notify_and_destroy(net, skb, n, classid,
850 if (new && !new->ops->attach)
851 atomic_inc(&new->refcnt);
852 dev->qdisc = new ? : &noop_qdisc;
854 if (new && new->ops->attach)
855 new->ops->attach(new);
857 notify_and_destroy(net, skb, n, classid, old, new);
860 if (dev->flags & IFF_UP)
863 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
866 if (cops && cops->graft) {
867 unsigned long cl = cops->get(parent, classid);
869 err = cops->graft(parent, cl, new, &old);
870 cops->put(parent, cl);
875 notify_and_destroy(net, skb, n, classid, old, new);
880 /* lockdep annotation is needed for ingress; egress gets it only for name */
881 static struct lock_class_key qdisc_tx_lock;
882 static struct lock_class_key qdisc_rx_lock;
885 Allocate and initialize new qdisc.
887 Parameters are passed via opt.
890 static struct Qdisc *qdisc_create(struct net_device *dev,
891 struct netdev_queue *dev_queue,
892 struct Qdisc *p, u32 parent, u32 handle,
893 struct nlattr **tca, int *errp)
896 struct nlattr *kind = tca[TCA_KIND];
898 struct Qdisc_ops *ops;
899 struct qdisc_size_table *stab;
901 ops = qdisc_lookup_ops(kind);
902 #ifdef CONFIG_MODULES
903 if (ops == NULL && kind != NULL) {
905 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
906 /* We dropped the RTNL semaphore in order to
907 * perform the module load. So, even if we
908 * succeeded in loading the module we have to
909 * tell the caller to replay the request. We
910 * indicate this using -EAGAIN.
911 * We replay the request because the device may
912 * go away in the mean time.
915 request_module("sch_%s", name);
917 ops = qdisc_lookup_ops(kind);
919 /* We will try again qdisc_lookup_ops,
920 * so don't keep a reference.
922 module_put(ops->owner);
934 sch = qdisc_alloc(dev_queue, ops);
940 sch->parent = parent;
942 if (handle == TC_H_INGRESS) {
943 sch->flags |= TCQ_F_INGRESS;
944 handle = TC_H_MAKE(TC_H_INGRESS, 0);
945 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
948 handle = qdisc_alloc_handle(dev);
953 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
954 if (!netif_is_multiqueue(dev))
955 sch->flags |= TCQ_F_ONETXQUEUE;
958 sch->handle = handle;
960 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
961 if (qdisc_is_percpu_stats(sch)) {
963 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
964 if (!sch->cpu_bstats)
967 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
968 if (!sch->cpu_qstats)
973 stab = qdisc_get_stab(tca[TCA_STAB]);
978 rcu_assign_pointer(sch->stab, stab);
984 if (sch->flags & TCQ_F_MQROOT)
987 if ((sch->parent != TC_H_ROOT) &&
988 !(sch->flags & TCQ_F_INGRESS) &&
989 (!p || !(p->flags & TCQ_F_MQROOT)))
990 running = qdisc_root_sleeping_running(sch);
992 running = &sch->running;
994 err = gen_new_estimator(&sch->bstats,
1004 qdisc_hash_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,
1065 qdisc_root_sleeping_running(sch),
1072 struct check_loop_arg {
1073 struct qdisc_walker w;
1078 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1079 struct qdisc_walker *w);
1081 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1083 struct check_loop_arg arg;
1085 if (q->ops->cl_ops == NULL)
1088 arg.w.stop = arg.w.skip = arg.w.count = 0;
1089 arg.w.fn = check_loop_fn;
1092 q->ops->cl_ops->walk(q, &arg.w);
1093 return arg.w.stop ? -ELOOP : 0;
1097 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1100 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1101 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1103 leaf = cops->leaf(q, cl);
1105 if (leaf == arg->p || arg->depth > 7)
1107 return check_loop(leaf, arg->p, arg->depth + 1);
1116 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1118 struct net *net = sock_net(skb->sk);
1119 struct tcmsg *tcm = nlmsg_data(n);
1120 struct nlattr *tca[TCA_MAX + 1];
1121 struct net_device *dev;
1123 struct Qdisc *q = NULL;
1124 struct Qdisc *p = NULL;
1127 if ((n->nlmsg_type != RTM_GETQDISC) &&
1128 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1131 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1135 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1139 clid = tcm->tcm_parent;
1141 if (clid != TC_H_ROOT) {
1142 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1143 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1146 q = qdisc_leaf(p, clid);
1147 } else if (dev_ingress_queue(dev)) {
1148 q = dev_ingress_queue(dev)->qdisc_sleeping;
1156 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1159 q = qdisc_lookup(dev, tcm->tcm_handle);
1164 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1167 if (n->nlmsg_type == RTM_DELQDISC) {
1172 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1176 qdisc_notify(net, skb, n, clid, NULL, q);
1182 * Create/change qdisc.
1185 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1187 struct net *net = sock_net(skb->sk);
1189 struct nlattr *tca[TCA_MAX + 1];
1190 struct net_device *dev;
1192 struct Qdisc *q, *p;
1195 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1199 /* Reinit, just in case something touches this. */
1200 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1204 tcm = nlmsg_data(n);
1205 clid = tcm->tcm_parent;
1208 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1214 if (clid != TC_H_ROOT) {
1215 if (clid != TC_H_INGRESS) {
1216 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1219 q = qdisc_leaf(p, clid);
1220 } else if (dev_ingress_queue_create(dev)) {
1221 q = dev_ingress_queue(dev)->qdisc_sleeping;
1227 /* It may be default qdisc, ignore it */
1228 if (q && q->handle == 0)
1231 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1232 if (tcm->tcm_handle) {
1233 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1235 if (TC_H_MIN(tcm->tcm_handle))
1237 q = qdisc_lookup(dev, tcm->tcm_handle);
1239 goto create_n_graft;
1240 if (n->nlmsg_flags & NLM_F_EXCL)
1242 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1245 (p && check_loop(q, p, 0)))
1247 atomic_inc(&q->refcnt);
1251 goto create_n_graft;
1253 /* This magic test requires explanation.
1255 * We know, that some child q is already
1256 * attached to this parent and have choice:
1257 * either to change it or to create/graft new one.
1259 * 1. We are allowed to create/graft only
1260 * if CREATE and REPLACE flags are set.
1262 * 2. If EXCL is set, requestor wanted to say,
1263 * that qdisc tcm_handle is not expected
1264 * to exist, so that we choose create/graft too.
1266 * 3. The last case is when no flags are set.
1267 * Alas, it is sort of hole in API, we
1268 * cannot decide what to do unambiguously.
1269 * For now we select create/graft, if
1270 * user gave KIND, which does not match existing.
1272 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1273 (n->nlmsg_flags & NLM_F_REPLACE) &&
1274 ((n->nlmsg_flags & NLM_F_EXCL) ||
1276 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1277 goto create_n_graft;
1281 if (!tcm->tcm_handle)
1283 q = qdisc_lookup(dev, tcm->tcm_handle);
1286 /* Change qdisc parameters */
1289 if (n->nlmsg_flags & NLM_F_EXCL)
1291 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1293 err = qdisc_change(q, tca);
1295 qdisc_notify(net, skb, n, clid, NULL, q);
1299 if (!(n->nlmsg_flags & NLM_F_CREATE))
1301 if (clid == TC_H_INGRESS) {
1302 if (dev_ingress_queue(dev))
1303 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1304 tcm->tcm_parent, tcm->tcm_parent,
1309 struct netdev_queue *dev_queue;
1311 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1312 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1314 dev_queue = p->dev_queue;
1316 dev_queue = netdev_get_tx_queue(dev, 0);
1318 q = qdisc_create(dev, dev_queue, p,
1319 tcm->tcm_parent, tcm->tcm_handle,
1329 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1339 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1340 u32 portid, u32 seq, u16 flags, int event)
1342 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
1343 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
1345 struct nlmsghdr *nlh;
1346 unsigned char *b = skb_tail_pointer(skb);
1348 struct qdisc_size_table *stab;
1352 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1354 goto out_nlmsg_trim;
1355 tcm = nlmsg_data(nlh);
1356 tcm->tcm_family = AF_UNSPEC;
1359 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1360 tcm->tcm_parent = clid;
1361 tcm->tcm_handle = q->handle;
1362 tcm->tcm_info = atomic_read(&q->refcnt);
1363 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1364 goto nla_put_failure;
1365 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1366 goto nla_put_failure;
1369 stab = rtnl_dereference(q->stab);
1370 if (stab && qdisc_dump_stab(skb, stab) < 0)
1371 goto nla_put_failure;
1373 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1374 NULL, &d, TCA_PAD) < 0)
1375 goto nla_put_failure;
1377 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1378 goto nla_put_failure;
1380 if (qdisc_is_percpu_stats(q)) {
1381 cpu_bstats = q->cpu_bstats;
1382 cpu_qstats = q->cpu_qstats;
1385 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
1386 &d, cpu_bstats, &q->bstats) < 0 ||
1387 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1388 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
1389 goto nla_put_failure;
1391 if (gnet_stats_finish_copy(&d) < 0)
1392 goto nla_put_failure;
1394 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1403 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1405 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1408 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1409 struct nlmsghdr *n, u32 clid,
1410 struct Qdisc *old, struct Qdisc *new)
1412 struct sk_buff *skb;
1413 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1415 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1419 if (old && !tc_qdisc_dump_ignore(old)) {
1420 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1421 0, RTM_DELQDISC) < 0)
1424 if (new && !tc_qdisc_dump_ignore(new)) {
1425 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1426 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1431 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1432 n->nlmsg_flags & NLM_F_ECHO);
1439 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1440 struct netlink_callback *cb,
1441 int *q_idx_p, int s_q_idx, bool recur)
1443 int ret = 0, q_idx = *q_idx_p;
1451 if (q_idx < s_q_idx) {
1454 if (!tc_qdisc_dump_ignore(q) &&
1455 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1456 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1462 /* If dumping singletons, there is no qdisc_dev(root) and the singleton
1463 * itself has already been dumped.
1465 * If we've already dumped the top-level (ingress) qdisc above and the global
1466 * qdisc hashtable, we don't want to hit it again
1468 if (!qdisc_dev(root) || !recur)
1471 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1472 if (q_idx < s_q_idx) {
1476 if (!tc_qdisc_dump_ignore(q) &&
1477 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1478 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1492 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1494 struct net *net = sock_net(skb->sk);
1497 struct net_device *dev;
1499 s_idx = cb->args[0];
1500 s_q_idx = q_idx = cb->args[1];
1504 for_each_netdev(net, dev) {
1505 struct netdev_queue *dev_queue;
1513 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1517 dev_queue = dev_ingress_queue(dev);
1519 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1520 &q_idx, s_q_idx, false) < 0)
1529 cb->args[1] = q_idx;
1536 /************************************************
1537 * Traffic classes manipulation. *
1538 ************************************************/
1542 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1544 struct net *net = sock_net(skb->sk);
1545 struct tcmsg *tcm = nlmsg_data(n);
1546 struct nlattr *tca[TCA_MAX + 1];
1547 struct net_device *dev;
1548 struct Qdisc *q = NULL;
1549 const struct Qdisc_class_ops *cops;
1550 unsigned long cl = 0;
1551 unsigned long new_cl;
1557 if ((n->nlmsg_type != RTM_GETTCLASS) &&
1558 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1561 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1565 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1570 parent == TC_H_UNSPEC - unspecified parent.
1571 parent == TC_H_ROOT - class is root, which has no parent.
1572 parent == X:0 - parent is root class.
1573 parent == X:Y - parent is a node in hierarchy.
1574 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1576 handle == 0:0 - generate handle from kernel pool.
1577 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1578 handle == X:Y - clear.
1579 handle == X:0 - root class.
1582 /* Step 1. Determine qdisc handle X:0 */
1584 portid = tcm->tcm_parent;
1585 clid = tcm->tcm_handle;
1586 qid = TC_H_MAJ(clid);
1588 if (portid != TC_H_ROOT) {
1589 u32 qid1 = TC_H_MAJ(portid);
1592 /* If both majors are known, they must be identical. */
1597 } else if (qid == 0)
1598 qid = dev->qdisc->handle;
1600 /* Now qid is genuine qdisc handle consistent
1601 * both with parent and child.
1603 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1606 portid = TC_H_MAKE(qid, portid);
1609 qid = dev->qdisc->handle;
1612 /* OK. Locate qdisc */
1613 q = qdisc_lookup(dev, qid);
1617 /* An check that it supports classes */
1618 cops = q->ops->cl_ops;
1622 /* Now try to get class */
1624 if (portid == TC_H_ROOT)
1627 clid = TC_H_MAKE(qid, clid);
1630 cl = cops->get(q, clid);
1634 if (n->nlmsg_type != RTM_NEWTCLASS ||
1635 !(n->nlmsg_flags & NLM_F_CREATE))
1638 switch (n->nlmsg_type) {
1641 if (n->nlmsg_flags & NLM_F_EXCL)
1647 err = cops->delete(q, cl);
1649 tclass_notify(net, skb, n, q, cl,
1653 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1664 err = cops->change(q, clid, portid, tca, &new_cl);
1666 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1676 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1678 u32 portid, u32 seq, u16 flags, int event)
1681 struct nlmsghdr *nlh;
1682 unsigned char *b = skb_tail_pointer(skb);
1684 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1687 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1689 goto out_nlmsg_trim;
1690 tcm = nlmsg_data(nlh);
1691 tcm->tcm_family = AF_UNSPEC;
1694 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1695 tcm->tcm_parent = q->handle;
1696 tcm->tcm_handle = q->handle;
1698 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1699 goto nla_put_failure;
1700 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1701 goto nla_put_failure;
1703 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1704 NULL, &d, TCA_PAD) < 0)
1705 goto nla_put_failure;
1707 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1708 goto nla_put_failure;
1710 if (gnet_stats_finish_copy(&d) < 0)
1711 goto nla_put_failure;
1713 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1722 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1723 struct nlmsghdr *n, struct Qdisc *q,
1724 unsigned long cl, int event)
1726 struct sk_buff *skb;
1727 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1729 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1733 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1738 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1739 n->nlmsg_flags & NLM_F_ECHO);
1742 struct qdisc_dump_args {
1743 struct qdisc_walker w;
1744 struct sk_buff *skb;
1745 struct netlink_callback *cb;
1748 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
1749 struct qdisc_walker *arg)
1751 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1753 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1754 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1758 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1759 struct tcmsg *tcm, struct netlink_callback *cb,
1762 struct qdisc_dump_args arg;
1764 if (tc_qdisc_dump_ignore(q) ||
1765 *t_p < s_t || !q->ops->cl_ops ||
1767 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1772 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1773 arg.w.fn = qdisc_class_dump;
1777 arg.w.skip = cb->args[1];
1779 q->ops->cl_ops->walk(q, &arg.w);
1780 cb->args[1] = arg.w.count;
1787 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1788 struct tcmsg *tcm, struct netlink_callback *cb,
1797 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1800 if (!qdisc_dev(root))
1803 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1804 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1811 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1813 struct tcmsg *tcm = nlmsg_data(cb->nlh);
1814 struct net *net = sock_net(skb->sk);
1815 struct netdev_queue *dev_queue;
1816 struct net_device *dev;
1819 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1821 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1828 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1831 dev_queue = dev_ingress_queue(dev);
1833 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1844 /* Main classifier routine: scans classifier chain attached
1845 * to this qdisc, (optionally) tests for protocol and asks
1846 * specific classifiers.
1848 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1849 struct tcf_result *res, bool compat_mode)
1851 __be16 protocol = tc_skb_protocol(skb);
1852 #ifdef CONFIG_NET_CLS_ACT
1853 const struct tcf_proto *old_tp = tp;
1858 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1861 if (tp->protocol != protocol &&
1862 tp->protocol != htons(ETH_P_ALL))
1865 err = tp->classify(skb, tp, res);
1866 #ifdef CONFIG_NET_CLS_ACT
1867 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode))
1874 return TC_ACT_UNSPEC; /* signal: continue lookup */
1875 #ifdef CONFIG_NET_CLS_ACT
1877 if (unlikely(limit++ >= MAX_REC_LOOP)) {
1878 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
1879 tp->q->ops->id, tp->prio & 0xffff,
1880 ntohs(tp->protocol));
1885 protocol = tc_skb_protocol(skb);
1889 EXPORT_SYMBOL(tc_classify);
1891 bool tcf_destroy(struct tcf_proto *tp, bool force)
1893 if (tp->ops->destroy(tp, force)) {
1894 module_put(tp->ops->owner);
1902 void tcf_destroy_chain(struct tcf_proto __rcu **fl)
1904 struct tcf_proto *tp;
1906 while ((tp = rtnl_dereference(*fl)) != NULL) {
1907 RCU_INIT_POINTER(*fl, tp->next);
1908 tcf_destroy(tp, true);
1911 EXPORT_SYMBOL(tcf_destroy_chain);
1913 #ifdef CONFIG_PROC_FS
1914 static int psched_show(struct seq_file *seq, void *v)
1916 seq_printf(seq, "%08x %08x %08x %08x\n",
1917 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1919 (u32)NSEC_PER_SEC / hrtimer_resolution);
1924 static int psched_open(struct inode *inode, struct file *file)
1926 return single_open(file, psched_show, NULL);
1929 static const struct file_operations psched_fops = {
1930 .owner = THIS_MODULE,
1931 .open = psched_open,
1933 .llseek = seq_lseek,
1934 .release = single_release,
1937 static int __net_init psched_net_init(struct net *net)
1939 struct proc_dir_entry *e;
1941 e = proc_create("psched", 0, net->proc_net, &psched_fops);
1948 static void __net_exit psched_net_exit(struct net *net)
1950 remove_proc_entry("psched", net->proc_net);
1953 static int __net_init psched_net_init(struct net *net)
1958 static void __net_exit psched_net_exit(struct net *net)
1963 static struct pernet_operations psched_net_ops = {
1964 .init = psched_net_init,
1965 .exit = psched_net_exit,
1968 static int __init pktsched_init(void)
1972 err = register_pernet_subsys(&psched_net_ops);
1974 pr_err("pktsched_init: "
1975 "cannot initialize per netns operations\n");
1979 register_qdisc(&pfifo_fast_ops);
1980 register_qdisc(&pfifo_qdisc_ops);
1981 register_qdisc(&bfifo_qdisc_ops);
1982 register_qdisc(&pfifo_head_drop_qdisc_ops);
1983 register_qdisc(&mq_qdisc_ops);
1984 register_qdisc(&noqueue_qdisc_ops);
1986 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1987 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1988 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
1990 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1991 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1992 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
1998 subsys_initcall(pktsched_init);