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.)
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 void qdisc_list_add(struct Qdisc *q)
276 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
277 struct Qdisc *root = qdisc_dev(q)->qdisc;
279 WARN_ON_ONCE(root == &noop_qdisc);
280 list_add_tail(&q->list, &root->list);
283 EXPORT_SYMBOL(qdisc_list_add);
285 void qdisc_list_del(struct Qdisc *q)
287 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
290 EXPORT_SYMBOL(qdisc_list_del);
292 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
296 q = qdisc_match_from_root(dev->qdisc, handle);
300 if (dev_ingress_queue(dev))
301 q = qdisc_match_from_root(
302 dev_ingress_queue(dev)->qdisc_sleeping,
308 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
312 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
316 cl = cops->get(p, classid);
320 leaf = cops->leaf(p, cl);
325 /* Find queueing discipline by name */
327 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
329 struct Qdisc_ops *q = NULL;
332 read_lock(&qdisc_mod_lock);
333 for (q = qdisc_base; q; q = q->next) {
334 if (nla_strcmp(kind, q->id) == 0) {
335 if (!try_module_get(q->owner))
340 read_unlock(&qdisc_mod_lock);
345 /* The linklayer setting were not transferred from iproute2, in older
346 * versions, and the rate tables lookup systems have been dropped in
347 * the kernel. To keep backward compatible with older iproute2 tc
348 * utils, we detect the linklayer setting by detecting if the rate
349 * table were modified.
351 * For linklayer ATM table entries, the rate table will be aligned to
352 * 48 bytes, thus some table entries will contain the same value. The
353 * mpu (min packet unit) is also encoded into the old rate table, thus
354 * starting from the mpu, we find low and high table entries for
355 * mapping this cell. If these entries contain the same value, when
356 * the rate tables have been modified for linklayer ATM.
358 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
359 * and then roundup to the next cell, calc the table entry one below,
362 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
364 int low = roundup(r->mpu, 48);
365 int high = roundup(low+1, 48);
366 int cell_low = low >> r->cell_log;
367 int cell_high = (high >> r->cell_log) - 1;
369 /* rtab is too inaccurate at rates > 100Mbit/s */
370 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
371 pr_debug("TC linklayer: Giving up ATM detection\n");
372 return TC_LINKLAYER_ETHERNET;
375 if ((cell_high > cell_low) && (cell_high < 256)
376 && (rtab[cell_low] == rtab[cell_high])) {
377 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
378 cell_low, cell_high, rtab[cell_high]);
379 return TC_LINKLAYER_ATM;
381 return TC_LINKLAYER_ETHERNET;
384 static struct qdisc_rate_table *qdisc_rtab_list;
386 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
388 struct qdisc_rate_table *rtab;
390 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
391 nla_len(tab) != TC_RTAB_SIZE)
394 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
395 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
396 !memcmp(&rtab->data, nla_data(tab), 1024)) {
402 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
406 memcpy(rtab->data, nla_data(tab), 1024);
407 if (r->linklayer == TC_LINKLAYER_UNAWARE)
408 r->linklayer = __detect_linklayer(r, rtab->data);
409 rtab->next = qdisc_rtab_list;
410 qdisc_rtab_list = rtab;
414 EXPORT_SYMBOL(qdisc_get_rtab);
416 void qdisc_put_rtab(struct qdisc_rate_table *tab)
418 struct qdisc_rate_table *rtab, **rtabp;
420 if (!tab || --tab->refcnt)
423 for (rtabp = &qdisc_rtab_list;
424 (rtab = *rtabp) != NULL;
425 rtabp = &rtab->next) {
433 EXPORT_SYMBOL(qdisc_put_rtab);
435 static LIST_HEAD(qdisc_stab_list);
436 static DEFINE_SPINLOCK(qdisc_stab_lock);
438 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
439 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
440 [TCA_STAB_DATA] = { .type = NLA_BINARY },
443 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
445 struct nlattr *tb[TCA_STAB_MAX + 1];
446 struct qdisc_size_table *stab;
447 struct tc_sizespec *s;
448 unsigned int tsize = 0;
452 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
455 if (!tb[TCA_STAB_BASE])
456 return ERR_PTR(-EINVAL);
458 s = nla_data(tb[TCA_STAB_BASE]);
461 if (!tb[TCA_STAB_DATA])
462 return ERR_PTR(-EINVAL);
463 tab = nla_data(tb[TCA_STAB_DATA]);
464 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
467 if (tsize != s->tsize || (!tab && tsize > 0))
468 return ERR_PTR(-EINVAL);
470 spin_lock(&qdisc_stab_lock);
472 list_for_each_entry(stab, &qdisc_stab_list, list) {
473 if (memcmp(&stab->szopts, s, sizeof(*s)))
475 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
478 spin_unlock(&qdisc_stab_lock);
482 spin_unlock(&qdisc_stab_lock);
484 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
486 return ERR_PTR(-ENOMEM);
491 memcpy(stab->data, tab, tsize * sizeof(u16));
493 spin_lock(&qdisc_stab_lock);
494 list_add_tail(&stab->list, &qdisc_stab_list);
495 spin_unlock(&qdisc_stab_lock);
500 static void stab_kfree_rcu(struct rcu_head *head)
502 kfree(container_of(head, struct qdisc_size_table, rcu));
505 void qdisc_put_stab(struct qdisc_size_table *tab)
510 spin_lock(&qdisc_stab_lock);
512 if (--tab->refcnt == 0) {
513 list_del(&tab->list);
514 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
517 spin_unlock(&qdisc_stab_lock);
519 EXPORT_SYMBOL(qdisc_put_stab);
521 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
525 nest = nla_nest_start(skb, TCA_STAB);
527 goto nla_put_failure;
528 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
529 goto nla_put_failure;
530 nla_nest_end(skb, nest);
538 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
542 pkt_len = skb->len + stab->szopts.overhead;
543 if (unlikely(!stab->szopts.tsize))
546 slot = pkt_len + stab->szopts.cell_align;
547 if (unlikely(slot < 0))
550 slot >>= stab->szopts.cell_log;
551 if (likely(slot < stab->szopts.tsize))
552 pkt_len = stab->data[slot];
554 pkt_len = stab->data[stab->szopts.tsize - 1] *
555 (slot / stab->szopts.tsize) +
556 stab->data[slot % stab->szopts.tsize];
558 pkt_len <<= stab->szopts.size_log;
560 if (unlikely(pkt_len < 1))
562 qdisc_skb_cb(skb)->pkt_len = pkt_len;
564 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
566 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
568 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
569 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
570 txt, qdisc->ops->id, qdisc->handle >> 16);
571 qdisc->flags |= TCQ_F_WARN_NONWC;
574 EXPORT_SYMBOL(qdisc_warn_nonwc);
576 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
578 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
581 qdisc_unthrottled(wd->qdisc);
582 __netif_schedule(qdisc_root(wd->qdisc));
584 return HRTIMER_NORESTART;
587 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
589 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
590 wd->timer.function = qdisc_watchdog;
593 EXPORT_SYMBOL(qdisc_watchdog_init);
595 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
597 if (test_bit(__QDISC_STATE_DEACTIVATED,
598 &qdisc_root_sleeping(wd->qdisc)->state))
601 qdisc_throttled(wd->qdisc);
603 hrtimer_start(&wd->timer,
604 ns_to_ktime(expires),
605 HRTIMER_MODE_ABS_PINNED);
607 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
609 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
611 hrtimer_cancel(&wd->timer);
612 qdisc_unthrottled(wd->qdisc);
614 EXPORT_SYMBOL(qdisc_watchdog_cancel);
616 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
618 unsigned int size = n * sizeof(struct hlist_head), i;
619 struct hlist_head *h;
621 if (size <= PAGE_SIZE)
622 h = kmalloc(size, GFP_KERNEL);
624 h = (struct hlist_head *)
625 __get_free_pages(GFP_KERNEL, get_order(size));
628 for (i = 0; i < n; i++)
629 INIT_HLIST_HEAD(&h[i]);
634 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
636 unsigned int size = n * sizeof(struct hlist_head);
638 if (size <= PAGE_SIZE)
641 free_pages((unsigned long)h, get_order(size));
644 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
646 struct Qdisc_class_common *cl;
647 struct hlist_node *next;
648 struct hlist_head *nhash, *ohash;
649 unsigned int nsize, nmask, osize;
652 /* Rehash when load factor exceeds 0.75 */
653 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
655 nsize = clhash->hashsize * 2;
657 nhash = qdisc_class_hash_alloc(nsize);
661 ohash = clhash->hash;
662 osize = clhash->hashsize;
665 for (i = 0; i < osize; i++) {
666 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
667 h = qdisc_class_hash(cl->classid, nmask);
668 hlist_add_head(&cl->hnode, &nhash[h]);
671 clhash->hash = nhash;
672 clhash->hashsize = nsize;
673 clhash->hashmask = nmask;
674 sch_tree_unlock(sch);
676 qdisc_class_hash_free(ohash, osize);
678 EXPORT_SYMBOL(qdisc_class_hash_grow);
680 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
682 unsigned int size = 4;
684 clhash->hash = qdisc_class_hash_alloc(size);
685 if (clhash->hash == NULL)
687 clhash->hashsize = size;
688 clhash->hashmask = size - 1;
689 clhash->hashelems = 0;
692 EXPORT_SYMBOL(qdisc_class_hash_init);
694 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
696 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
698 EXPORT_SYMBOL(qdisc_class_hash_destroy);
700 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
701 struct Qdisc_class_common *cl)
705 INIT_HLIST_NODE(&cl->hnode);
706 h = qdisc_class_hash(cl->classid, clhash->hashmask);
707 hlist_add_head(&cl->hnode, &clhash->hash[h]);
710 EXPORT_SYMBOL(qdisc_class_hash_insert);
712 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
713 struct Qdisc_class_common *cl)
715 hlist_del(&cl->hnode);
718 EXPORT_SYMBOL(qdisc_class_hash_remove);
720 /* Allocate an unique handle from space managed by kernel
721 * Possible range is [8000-FFFF]:0000 (0x8000 values)
723 static u32 qdisc_alloc_handle(struct net_device *dev)
726 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
729 autohandle += TC_H_MAKE(0x10000U, 0);
730 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
731 autohandle = TC_H_MAKE(0x80000000U, 0);
732 if (!qdisc_lookup(dev, autohandle))
740 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
742 const struct Qdisc_class_ops *cops;
749 drops = max_t(int, n, 0);
750 while ((parentid = sch->parent)) {
751 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
754 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
756 WARN_ON(parentid != TC_H_ROOT);
759 cops = sch->ops->cl_ops;
760 if (cops->qlen_notify) {
761 cl = cops->get(sch, parentid);
762 cops->qlen_notify(sch, cl);
766 __qdisc_qstats_drop(sch, drops);
769 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
771 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
772 struct nlmsghdr *n, u32 clid,
773 struct Qdisc *old, struct Qdisc *new)
776 qdisc_notify(net, skb, n, clid, old, new);
782 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
785 * When appropriate send a netlink notification using 'skb'
788 * On success, destroy old qdisc.
791 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
792 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
793 struct Qdisc *new, struct Qdisc *old)
795 struct Qdisc *q = old;
796 struct net *net = dev_net(dev);
799 if (parent == NULL) {
800 unsigned int i, num_q, ingress;
803 num_q = dev->num_tx_queues;
804 if ((q && q->flags & TCQ_F_INGRESS) ||
805 (new && new->flags & TCQ_F_INGRESS)) {
808 if (!dev_ingress_queue(dev))
812 if (dev->flags & IFF_UP)
815 if (new && new->ops->attach) {
816 new->ops->attach(new);
820 for (i = 0; i < num_q; i++) {
821 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
824 dev_queue = netdev_get_tx_queue(dev, i);
826 old = dev_graft_qdisc(dev_queue, new);
828 atomic_inc(&new->refcnt);
835 notify_and_destroy(net, skb, n, classid,
837 if (new && !new->ops->attach)
838 atomic_inc(&new->refcnt);
839 dev->qdisc = new ? : &noop_qdisc;
841 notify_and_destroy(net, skb, n, classid, old, new);
844 if (dev->flags & IFF_UP)
847 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
850 if (cops && cops->graft) {
851 unsigned long cl = cops->get(parent, classid);
853 err = cops->graft(parent, cl, new, &old);
854 cops->put(parent, cl);
859 notify_and_destroy(net, skb, n, classid, old, new);
864 /* lockdep annotation is needed for ingress; egress gets it only for name */
865 static struct lock_class_key qdisc_tx_lock;
866 static struct lock_class_key qdisc_rx_lock;
869 Allocate and initialize new qdisc.
871 Parameters are passed via opt.
874 static struct Qdisc *
875 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
876 struct Qdisc *p, u32 parent, u32 handle,
877 struct nlattr **tca, int *errp)
880 struct nlattr *kind = tca[TCA_KIND];
882 struct Qdisc_ops *ops;
883 struct qdisc_size_table *stab;
885 ops = qdisc_lookup_ops(kind);
886 #ifdef CONFIG_MODULES
887 if (ops == NULL && kind != NULL) {
889 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
890 /* We dropped the RTNL semaphore in order to
891 * perform the module load. So, even if we
892 * succeeded in loading the module we have to
893 * tell the caller to replay the request. We
894 * indicate this using -EAGAIN.
895 * We replay the request because the device may
896 * go away in the mean time.
899 request_module("sch_%s", name);
901 ops = qdisc_lookup_ops(kind);
903 /* We will try again qdisc_lookup_ops,
904 * so don't keep a reference.
906 module_put(ops->owner);
918 sch = qdisc_alloc(dev_queue, ops);
924 sch->parent = parent;
926 if (handle == TC_H_INGRESS) {
927 sch->flags |= TCQ_F_INGRESS;
928 handle = TC_H_MAKE(TC_H_INGRESS, 0);
929 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
932 handle = qdisc_alloc_handle(dev);
937 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
938 if (!netif_is_multiqueue(dev))
939 sch->flags |= TCQ_F_ONETXQUEUE;
942 sch->handle = handle;
944 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
945 if (qdisc_is_percpu_stats(sch)) {
947 alloc_percpu(struct gnet_stats_basic_cpu);
948 if (!sch->cpu_bstats)
951 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
952 if (!sch->cpu_qstats)
957 stab = qdisc_get_stab(tca[TCA_STAB]);
962 rcu_assign_pointer(sch->stab, stab);
965 spinlock_t *root_lock;
968 if (sch->flags & TCQ_F_MQROOT)
971 if ((sch->parent != TC_H_ROOT) &&
972 !(sch->flags & TCQ_F_INGRESS) &&
973 (!p || !(p->flags & TCQ_F_MQROOT)))
974 root_lock = qdisc_root_sleeping_lock(sch);
976 root_lock = qdisc_lock(sch);
978 err = gen_new_estimator(&sch->bstats,
993 kfree((char *) sch - sch->padded);
995 module_put(ops->owner);
1001 free_percpu(sch->cpu_bstats);
1002 free_percpu(sch->cpu_qstats);
1004 * Any broken qdiscs that would require a ops->reset() here?
1005 * The qdisc was never in action so it shouldn't be necessary.
1007 qdisc_put_stab(rtnl_dereference(sch->stab));
1013 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
1015 struct qdisc_size_table *ostab, *stab = NULL;
1018 if (tca[TCA_OPTIONS]) {
1019 if (sch->ops->change == NULL)
1021 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1026 if (tca[TCA_STAB]) {
1027 stab = qdisc_get_stab(tca[TCA_STAB]);
1029 return PTR_ERR(stab);
1032 ostab = rtnl_dereference(sch->stab);
1033 rcu_assign_pointer(sch->stab, stab);
1034 qdisc_put_stab(ostab);
1036 if (tca[TCA_RATE]) {
1037 /* NB: ignores errors from replace_estimator
1038 because change can't be undone. */
1039 if (sch->flags & TCQ_F_MQROOT)
1041 gen_replace_estimator(&sch->bstats,
1044 qdisc_root_sleeping_lock(sch),
1051 struct check_loop_arg {
1052 struct qdisc_walker w;
1057 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
1059 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1061 struct check_loop_arg arg;
1063 if (q->ops->cl_ops == NULL)
1066 arg.w.stop = arg.w.skip = arg.w.count = 0;
1067 arg.w.fn = check_loop_fn;
1070 q->ops->cl_ops->walk(q, &arg.w);
1071 return arg.w.stop ? -ELOOP : 0;
1075 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1078 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1079 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1081 leaf = cops->leaf(q, cl);
1083 if (leaf == arg->p || arg->depth > 7)
1085 return check_loop(leaf, arg->p, arg->depth + 1);
1094 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1096 struct net *net = sock_net(skb->sk);
1097 struct tcmsg *tcm = nlmsg_data(n);
1098 struct nlattr *tca[TCA_MAX + 1];
1099 struct net_device *dev;
1101 struct Qdisc *q = NULL;
1102 struct Qdisc *p = NULL;
1105 if ((n->nlmsg_type != RTM_GETQDISC) &&
1106 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1109 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1113 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1117 clid = tcm->tcm_parent;
1119 if (clid != TC_H_ROOT) {
1120 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1121 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1124 q = qdisc_leaf(p, clid);
1125 } else if (dev_ingress_queue(dev)) {
1126 q = dev_ingress_queue(dev)->qdisc_sleeping;
1134 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1137 q = qdisc_lookup(dev, tcm->tcm_handle);
1142 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1145 if (n->nlmsg_type == RTM_DELQDISC) {
1150 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1154 qdisc_notify(net, skb, n, clid, NULL, q);
1160 * Create/change qdisc.
1163 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1165 struct net *net = sock_net(skb->sk);
1167 struct nlattr *tca[TCA_MAX + 1];
1168 struct net_device *dev;
1170 struct Qdisc *q, *p;
1173 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1177 /* Reinit, just in case something touches this. */
1178 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1182 tcm = nlmsg_data(n);
1183 clid = tcm->tcm_parent;
1186 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1192 if (clid != TC_H_ROOT) {
1193 if (clid != TC_H_INGRESS) {
1194 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1197 q = qdisc_leaf(p, clid);
1198 } else if (dev_ingress_queue_create(dev)) {
1199 q = dev_ingress_queue(dev)->qdisc_sleeping;
1205 /* It may be default qdisc, ignore it */
1206 if (q && q->handle == 0)
1209 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1210 if (tcm->tcm_handle) {
1211 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1213 if (TC_H_MIN(tcm->tcm_handle))
1215 q = qdisc_lookup(dev, tcm->tcm_handle);
1217 goto create_n_graft;
1218 if (n->nlmsg_flags & NLM_F_EXCL)
1220 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1223 (p && check_loop(q, p, 0)))
1225 atomic_inc(&q->refcnt);
1229 goto create_n_graft;
1231 /* This magic test requires explanation.
1233 * We know, that some child q is already
1234 * attached to this parent and have choice:
1235 * either to change it or to create/graft new one.
1237 * 1. We are allowed to create/graft only
1238 * if CREATE and REPLACE flags are set.
1240 * 2. If EXCL is set, requestor wanted to say,
1241 * that qdisc tcm_handle is not expected
1242 * to exist, so that we choose create/graft too.
1244 * 3. The last case is when no flags are set.
1245 * Alas, it is sort of hole in API, we
1246 * cannot decide what to do unambiguously.
1247 * For now we select create/graft, if
1248 * user gave KIND, which does not match existing.
1250 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1251 (n->nlmsg_flags & NLM_F_REPLACE) &&
1252 ((n->nlmsg_flags & NLM_F_EXCL) ||
1254 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1255 goto create_n_graft;
1259 if (!tcm->tcm_handle)
1261 q = qdisc_lookup(dev, tcm->tcm_handle);
1264 /* Change qdisc parameters */
1267 if (n->nlmsg_flags & NLM_F_EXCL)
1269 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1271 err = qdisc_change(q, tca);
1273 qdisc_notify(net, skb, n, clid, NULL, q);
1277 if (!(n->nlmsg_flags & NLM_F_CREATE))
1279 if (clid == TC_H_INGRESS) {
1280 if (dev_ingress_queue(dev))
1281 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1282 tcm->tcm_parent, tcm->tcm_parent,
1287 struct netdev_queue *dev_queue;
1289 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1290 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1292 dev_queue = p->dev_queue;
1294 dev_queue = netdev_get_tx_queue(dev, 0);
1296 q = qdisc_create(dev, dev_queue, p,
1297 tcm->tcm_parent, tcm->tcm_handle,
1307 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1317 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1318 u32 portid, u32 seq, u16 flags, int event)
1320 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
1321 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
1323 struct nlmsghdr *nlh;
1324 unsigned char *b = skb_tail_pointer(skb);
1326 struct qdisc_size_table *stab;
1330 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1332 goto out_nlmsg_trim;
1333 tcm = nlmsg_data(nlh);
1334 tcm->tcm_family = AF_UNSPEC;
1337 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1338 tcm->tcm_parent = clid;
1339 tcm->tcm_handle = q->handle;
1340 tcm->tcm_info = atomic_read(&q->refcnt);
1341 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1342 goto nla_put_failure;
1343 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1344 goto nla_put_failure;
1347 stab = rtnl_dereference(q->stab);
1348 if (stab && qdisc_dump_stab(skb, stab) < 0)
1349 goto nla_put_failure;
1351 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1352 qdisc_root_sleeping_lock(q), &d) < 0)
1353 goto nla_put_failure;
1355 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1356 goto nla_put_failure;
1358 if (qdisc_is_percpu_stats(q)) {
1359 cpu_bstats = q->cpu_bstats;
1360 cpu_qstats = q->cpu_qstats;
1363 if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats) < 0 ||
1364 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1365 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
1366 goto nla_put_failure;
1368 if (gnet_stats_finish_copy(&d) < 0)
1369 goto nla_put_failure;
1371 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1380 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1382 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1385 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1386 struct nlmsghdr *n, u32 clid,
1387 struct Qdisc *old, struct Qdisc *new)
1389 struct sk_buff *skb;
1390 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1392 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1396 if (old && !tc_qdisc_dump_ignore(old)) {
1397 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1398 0, RTM_DELQDISC) < 0)
1401 if (new && !tc_qdisc_dump_ignore(new)) {
1402 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1403 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1408 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1409 n->nlmsg_flags & NLM_F_ECHO);
1416 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1417 struct netlink_callback *cb,
1418 int *q_idx_p, int s_q_idx)
1420 int ret = 0, q_idx = *q_idx_p;
1427 if (q_idx < s_q_idx) {
1430 if (!tc_qdisc_dump_ignore(q) &&
1431 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1432 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1436 list_for_each_entry(q, &root->list, list) {
1437 if (q_idx < s_q_idx) {
1441 if (!tc_qdisc_dump_ignore(q) &&
1442 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1443 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1456 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1458 struct net *net = sock_net(skb->sk);
1461 struct net_device *dev;
1463 s_idx = cb->args[0];
1464 s_q_idx = q_idx = cb->args[1];
1468 for_each_netdev(net, dev) {
1469 struct netdev_queue *dev_queue;
1477 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1480 dev_queue = dev_ingress_queue(dev);
1482 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1483 &q_idx, s_q_idx) < 0)
1492 cb->args[1] = q_idx;
1499 /************************************************
1500 * Traffic classes manipulation. *
1501 ************************************************/
1505 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1507 struct net *net = sock_net(skb->sk);
1508 struct tcmsg *tcm = nlmsg_data(n);
1509 struct nlattr *tca[TCA_MAX + 1];
1510 struct net_device *dev;
1511 struct Qdisc *q = NULL;
1512 const struct Qdisc_class_ops *cops;
1513 unsigned long cl = 0;
1514 unsigned long new_cl;
1520 if ((n->nlmsg_type != RTM_GETTCLASS) &&
1521 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1524 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1528 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1533 parent == TC_H_UNSPEC - unspecified parent.
1534 parent == TC_H_ROOT - class is root, which has no parent.
1535 parent == X:0 - parent is root class.
1536 parent == X:Y - parent is a node in hierarchy.
1537 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1539 handle == 0:0 - generate handle from kernel pool.
1540 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1541 handle == X:Y - clear.
1542 handle == X:0 - root class.
1545 /* Step 1. Determine qdisc handle X:0 */
1547 portid = tcm->tcm_parent;
1548 clid = tcm->tcm_handle;
1549 qid = TC_H_MAJ(clid);
1551 if (portid != TC_H_ROOT) {
1552 u32 qid1 = TC_H_MAJ(portid);
1555 /* If both majors are known, they must be identical. */
1560 } else if (qid == 0)
1561 qid = dev->qdisc->handle;
1563 /* Now qid is genuine qdisc handle consistent
1564 * both with parent and child.
1566 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1569 portid = TC_H_MAKE(qid, portid);
1572 qid = dev->qdisc->handle;
1575 /* OK. Locate qdisc */
1576 q = qdisc_lookup(dev, qid);
1580 /* An check that it supports classes */
1581 cops = q->ops->cl_ops;
1585 /* Now try to get class */
1587 if (portid == TC_H_ROOT)
1590 clid = TC_H_MAKE(qid, clid);
1593 cl = cops->get(q, clid);
1597 if (n->nlmsg_type != RTM_NEWTCLASS ||
1598 !(n->nlmsg_flags & NLM_F_CREATE))
1601 switch (n->nlmsg_type) {
1604 if (n->nlmsg_flags & NLM_F_EXCL)
1610 err = cops->delete(q, cl);
1612 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1615 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1626 err = cops->change(q, clid, portid, tca, &new_cl);
1628 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1638 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1640 u32 portid, u32 seq, u16 flags, int event)
1643 struct nlmsghdr *nlh;
1644 unsigned char *b = skb_tail_pointer(skb);
1646 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1649 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1651 goto out_nlmsg_trim;
1652 tcm = nlmsg_data(nlh);
1653 tcm->tcm_family = AF_UNSPEC;
1656 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1657 tcm->tcm_parent = q->handle;
1658 tcm->tcm_handle = q->handle;
1660 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1661 goto nla_put_failure;
1662 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1663 goto nla_put_failure;
1665 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1666 qdisc_root_sleeping_lock(q), &d) < 0)
1667 goto nla_put_failure;
1669 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1670 goto nla_put_failure;
1672 if (gnet_stats_finish_copy(&d) < 0)
1673 goto nla_put_failure;
1675 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1684 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1685 struct nlmsghdr *n, struct Qdisc *q,
1686 unsigned long cl, int event)
1688 struct sk_buff *skb;
1689 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1691 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1695 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1700 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1701 n->nlmsg_flags & NLM_F_ECHO);
1704 struct qdisc_dump_args {
1705 struct qdisc_walker w;
1706 struct sk_buff *skb;
1707 struct netlink_callback *cb;
1710 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1712 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1714 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1715 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1718 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1719 struct tcmsg *tcm, struct netlink_callback *cb,
1722 struct qdisc_dump_args arg;
1724 if (tc_qdisc_dump_ignore(q) ||
1725 *t_p < s_t || !q->ops->cl_ops ||
1727 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1732 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1733 arg.w.fn = qdisc_class_dump;
1737 arg.w.skip = cb->args[1];
1739 q->ops->cl_ops->walk(q, &arg.w);
1740 cb->args[1] = arg.w.count;
1747 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1748 struct tcmsg *tcm, struct netlink_callback *cb,
1756 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1759 list_for_each_entry(q, &root->list, list) {
1760 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1767 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1769 struct tcmsg *tcm = nlmsg_data(cb->nlh);
1770 struct net *net = sock_net(skb->sk);
1771 struct netdev_queue *dev_queue;
1772 struct net_device *dev;
1775 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1777 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1784 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1787 dev_queue = dev_ingress_queue(dev);
1789 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1800 /* Main classifier routine: scans classifier chain attached
1801 * to this qdisc, (optionally) tests for protocol and asks
1802 * specific classifiers.
1804 int tc_classify_compat(struct sk_buff *skb, const struct tcf_proto *tp,
1805 struct tcf_result *res)
1807 __be16 protocol = skb->protocol;
1810 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1811 if (tp->protocol != protocol &&
1812 tp->protocol != htons(ETH_P_ALL))
1814 err = tp->classify(skb, tp, res);
1817 #ifdef CONFIG_NET_CLS_ACT
1818 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1819 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1826 EXPORT_SYMBOL(tc_classify_compat);
1828 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1829 struct tcf_result *res)
1832 #ifdef CONFIG_NET_CLS_ACT
1833 const struct tcf_proto *otp = tp;
1837 err = tc_classify_compat(skb, tp, res);
1838 #ifdef CONFIG_NET_CLS_ACT
1839 if (err == TC_ACT_RECLASSIFY) {
1840 u32 verd = G_TC_VERD(skb->tc_verd);
1843 if (verd++ >= MAX_REC_LOOP) {
1844 net_notice_ratelimited("%s: packet reclassify loop rule prio %u protocol %02x\n",
1847 ntohs(tp->protocol));
1850 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1856 EXPORT_SYMBOL(tc_classify);
1858 void tcf_destroy(struct tcf_proto *tp)
1860 tp->ops->destroy(tp);
1861 module_put(tp->ops->owner);
1865 void tcf_destroy_chain(struct tcf_proto __rcu **fl)
1867 struct tcf_proto *tp;
1869 while ((tp = rtnl_dereference(*fl)) != NULL) {
1870 RCU_INIT_POINTER(*fl, tp->next);
1874 EXPORT_SYMBOL(tcf_destroy_chain);
1876 #ifdef CONFIG_PROC_FS
1877 static int psched_show(struct seq_file *seq, void *v)
1881 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1882 seq_printf(seq, "%08x %08x %08x %08x\n",
1883 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1885 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1890 static int psched_open(struct inode *inode, struct file *file)
1892 return single_open(file, psched_show, NULL);
1895 static const struct file_operations psched_fops = {
1896 .owner = THIS_MODULE,
1897 .open = psched_open,
1899 .llseek = seq_lseek,
1900 .release = single_release,
1903 static int __net_init psched_net_init(struct net *net)
1905 struct proc_dir_entry *e;
1907 e = proc_create("psched", 0, net->proc_net, &psched_fops);
1914 static void __net_exit psched_net_exit(struct net *net)
1916 remove_proc_entry("psched", net->proc_net);
1919 static int __net_init psched_net_init(struct net *net)
1924 static void __net_exit psched_net_exit(struct net *net)
1929 static struct pernet_operations psched_net_ops = {
1930 .init = psched_net_init,
1931 .exit = psched_net_exit,
1934 static int __init pktsched_init(void)
1938 err = register_pernet_subsys(&psched_net_ops);
1940 pr_err("pktsched_init: "
1941 "cannot initialize per netns operations\n");
1945 register_qdisc(&pfifo_fast_ops);
1946 register_qdisc(&pfifo_qdisc_ops);
1947 register_qdisc(&bfifo_qdisc_ops);
1948 register_qdisc(&pfifo_head_drop_qdisc_ops);
1949 register_qdisc(&mq_qdisc_ops);
1951 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1952 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1953 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1954 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1955 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1956 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1961 subsys_initcall(pktsched_init);