2 * net/sched/sch_api.c Packet scheduler API.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
13 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired.
14 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
15 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support
18 #include <linux/module.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/string.h>
22 #include <linux/errno.h>
23 #include <linux/skbuff.h>
24 #include <linux/init.h>
25 #include <linux/proc_fs.h>
26 #include <linux/seq_file.h>
27 #include <linux/kmod.h>
28 #include <linux/list.h>
29 #include <linux/hrtimer.h>
30 #include <linux/lockdep.h>
31 #include <linux/slab.h>
33 #include <net/net_namespace.h>
35 #include <net/netlink.h>
36 #include <net/pkt_sched.h>
38 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
39 struct nlmsghdr *n, u32 clid,
40 struct Qdisc *old, struct Qdisc *new);
41 static int tclass_notify(struct net *net, struct sk_buff *oskb,
42 struct nlmsghdr *n, struct Qdisc *q,
43 unsigned long cl, int event);
50 This file consists of two interrelated parts:
52 1. queueing disciplines manager frontend.
53 2. traffic classes manager frontend.
55 Generally, queueing discipline ("qdisc") is a black box,
56 which is able to enqueue packets and to dequeue them (when
57 device is ready to send something) in order and at times
58 determined by algorithm hidden in it.
60 qdisc's are divided to two categories:
61 - "queues", which have no internal structure visible from outside.
62 - "schedulers", which split all the packets to "traffic classes",
63 using "packet classifiers" (look at cls_api.c)
65 In turn, classes may have child qdiscs (as rule, queues)
66 attached to them etc. etc. etc.
68 The goal of the routines in this file is to translate
69 information supplied by user in the form of handles
70 to more intelligible for kernel form, to make some sanity
71 checks and part of work, which is common to all qdiscs
72 and to provide rtnetlink notifications.
74 All real intelligent work is done inside qdisc modules.
78 Every discipline has two major routines: enqueue and dequeue.
82 dequeue usually returns a skb to send. It is allowed to return NULL,
83 but it does not mean that queue is empty, it just means that
84 discipline does not want to send anything this time.
85 Queue is really empty if q->q.qlen == 0.
86 For complicated disciplines with multiple queues q->q is not
87 real packet queue, but however q->q.qlen must be valid.
91 enqueue returns 0, if packet was enqueued successfully.
92 If packet (this one or another one) was dropped, it returns
94 NET_XMIT_DROP - this packet dropped
95 Expected action: do not backoff, but wait until queue will clear.
96 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
97 Expected action: backoff or ignore
98 NET_XMIT_POLICED - dropped by police.
99 Expected action: backoff or error to real-time apps.
105 like dequeue but without removing a packet from the queue
109 returns qdisc to initial state: purge all buffers, clear all
110 timers, counters (except for statistics) etc.
114 initializes newly created qdisc.
118 destroys resources allocated by init and during lifetime of qdisc.
122 changes qdisc parameters.
125 /* Protects list of registered TC modules. It is pure SMP lock. */
126 static DEFINE_RWLOCK(qdisc_mod_lock);
129 /************************************************
130 * Queueing disciplines manipulation. *
131 ************************************************/
134 /* The list of all installed queueing disciplines. */
136 static struct Qdisc_ops *qdisc_base;
138 /* Register/uregister queueing discipline */
140 int register_qdisc(struct Qdisc_ops *qops)
142 struct Qdisc_ops *q, **qp;
145 write_lock(&qdisc_mod_lock);
146 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
147 if (!strcmp(qops->id, q->id))
150 if (qops->enqueue == NULL)
151 qops->enqueue = noop_qdisc_ops.enqueue;
152 if (qops->peek == NULL) {
153 if (qops->dequeue == NULL)
154 qops->peek = noop_qdisc_ops.peek;
158 if (qops->dequeue == NULL)
159 qops->dequeue = noop_qdisc_ops.dequeue;
162 const struct Qdisc_class_ops *cops = qops->cl_ops;
164 if (!(cops->get && cops->put && cops->walk && cops->leaf))
167 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf))
175 write_unlock(&qdisc_mod_lock);
182 EXPORT_SYMBOL(register_qdisc);
184 int unregister_qdisc(struct Qdisc_ops *qops)
186 struct Qdisc_ops *q, **qp;
189 write_lock(&qdisc_mod_lock);
190 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
198 write_unlock(&qdisc_mod_lock);
201 EXPORT_SYMBOL(unregister_qdisc);
203 /* We know handle. Find qdisc among all qdisc's attached to device
204 (root qdisc, all its children, children of children etc.)
207 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
211 if (!(root->flags & TCQ_F_BUILTIN) &&
212 root->handle == handle)
215 list_for_each_entry(q, &root->list, list) {
216 if (q->handle == handle)
222 static void qdisc_list_add(struct Qdisc *q)
224 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
225 list_add_tail(&q->list, &qdisc_dev(q)->qdisc->list);
228 void qdisc_list_del(struct Qdisc *q)
230 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
233 EXPORT_SYMBOL(qdisc_list_del);
235 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
239 q = qdisc_match_from_root(dev->qdisc, handle);
243 if (dev_ingress_queue(dev))
244 q = qdisc_match_from_root(
245 dev_ingress_queue(dev)->qdisc_sleeping,
251 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
255 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
259 cl = cops->get(p, classid);
263 leaf = cops->leaf(p, cl);
268 /* Find queueing discipline by name */
270 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
272 struct Qdisc_ops *q = NULL;
275 read_lock(&qdisc_mod_lock);
276 for (q = qdisc_base; q; q = q->next) {
277 if (nla_strcmp(kind, q->id) == 0) {
278 if (!try_module_get(q->owner))
283 read_unlock(&qdisc_mod_lock);
288 static struct qdisc_rate_table *qdisc_rtab_list;
290 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
292 struct qdisc_rate_table *rtab;
294 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
295 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
301 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
302 nla_len(tab) != TC_RTAB_SIZE)
305 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
309 memcpy(rtab->data, nla_data(tab), 1024);
310 rtab->next = qdisc_rtab_list;
311 qdisc_rtab_list = rtab;
315 EXPORT_SYMBOL(qdisc_get_rtab);
317 void qdisc_put_rtab(struct qdisc_rate_table *tab)
319 struct qdisc_rate_table *rtab, **rtabp;
321 if (!tab || --tab->refcnt)
324 for (rtabp = &qdisc_rtab_list;
325 (rtab = *rtabp) != NULL;
326 rtabp = &rtab->next) {
334 EXPORT_SYMBOL(qdisc_put_rtab);
336 static LIST_HEAD(qdisc_stab_list);
337 static DEFINE_SPINLOCK(qdisc_stab_lock);
339 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
340 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
341 [TCA_STAB_DATA] = { .type = NLA_BINARY },
344 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
346 struct nlattr *tb[TCA_STAB_MAX + 1];
347 struct qdisc_size_table *stab;
348 struct tc_sizespec *s;
349 unsigned int tsize = 0;
353 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
356 if (!tb[TCA_STAB_BASE])
357 return ERR_PTR(-EINVAL);
359 s = nla_data(tb[TCA_STAB_BASE]);
362 if (!tb[TCA_STAB_DATA])
363 return ERR_PTR(-EINVAL);
364 tab = nla_data(tb[TCA_STAB_DATA]);
365 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
368 if (tsize != s->tsize || (!tab && tsize > 0))
369 return ERR_PTR(-EINVAL);
371 spin_lock(&qdisc_stab_lock);
373 list_for_each_entry(stab, &qdisc_stab_list, list) {
374 if (memcmp(&stab->szopts, s, sizeof(*s)))
376 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
379 spin_unlock(&qdisc_stab_lock);
383 spin_unlock(&qdisc_stab_lock);
385 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
387 return ERR_PTR(-ENOMEM);
392 memcpy(stab->data, tab, tsize * sizeof(u16));
394 spin_lock(&qdisc_stab_lock);
395 list_add_tail(&stab->list, &qdisc_stab_list);
396 spin_unlock(&qdisc_stab_lock);
401 static void stab_kfree_rcu(struct rcu_head *head)
403 kfree(container_of(head, struct qdisc_size_table, rcu));
406 void qdisc_put_stab(struct qdisc_size_table *tab)
411 spin_lock(&qdisc_stab_lock);
413 if (--tab->refcnt == 0) {
414 list_del(&tab->list);
415 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
418 spin_unlock(&qdisc_stab_lock);
420 EXPORT_SYMBOL(qdisc_put_stab);
422 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
426 nest = nla_nest_start(skb, TCA_STAB);
428 goto nla_put_failure;
429 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
430 goto nla_put_failure;
431 nla_nest_end(skb, nest);
439 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
443 pkt_len = skb->len + stab->szopts.overhead;
444 if (unlikely(!stab->szopts.tsize))
447 slot = pkt_len + stab->szopts.cell_align;
448 if (unlikely(slot < 0))
451 slot >>= stab->szopts.cell_log;
452 if (likely(slot < stab->szopts.tsize))
453 pkt_len = stab->data[slot];
455 pkt_len = stab->data[stab->szopts.tsize - 1] *
456 (slot / stab->szopts.tsize) +
457 stab->data[slot % stab->szopts.tsize];
459 pkt_len <<= stab->szopts.size_log;
461 if (unlikely(pkt_len < 1))
463 qdisc_skb_cb(skb)->pkt_len = pkt_len;
465 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
467 void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
469 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
470 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
471 txt, qdisc->ops->id, qdisc->handle >> 16);
472 qdisc->flags |= TCQ_F_WARN_NONWC;
475 EXPORT_SYMBOL(qdisc_warn_nonwc);
477 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
479 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
482 qdisc_unthrottled(wd->qdisc);
483 __netif_schedule(qdisc_root(wd->qdisc));
485 return HRTIMER_NORESTART;
488 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
490 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
491 wd->timer.function = qdisc_watchdog;
494 EXPORT_SYMBOL(qdisc_watchdog_init);
496 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
498 if (test_bit(__QDISC_STATE_DEACTIVATED,
499 &qdisc_root_sleeping(wd->qdisc)->state))
502 qdisc_throttled(wd->qdisc);
504 hrtimer_start(&wd->timer,
505 ns_to_ktime(expires),
508 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
510 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
512 hrtimer_cancel(&wd->timer);
513 qdisc_unthrottled(wd->qdisc);
515 EXPORT_SYMBOL(qdisc_watchdog_cancel);
517 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
519 unsigned int size = n * sizeof(struct hlist_head), i;
520 struct hlist_head *h;
522 if (size <= PAGE_SIZE)
523 h = kmalloc(size, GFP_KERNEL);
525 h = (struct hlist_head *)
526 __get_free_pages(GFP_KERNEL, get_order(size));
529 for (i = 0; i < n; i++)
530 INIT_HLIST_HEAD(&h[i]);
535 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
537 unsigned int size = n * sizeof(struct hlist_head);
539 if (size <= PAGE_SIZE)
542 free_pages((unsigned long)h, get_order(size));
545 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
547 struct Qdisc_class_common *cl;
548 struct hlist_node *next;
549 struct hlist_head *nhash, *ohash;
550 unsigned int nsize, nmask, osize;
553 /* Rehash when load factor exceeds 0.75 */
554 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
556 nsize = clhash->hashsize * 2;
558 nhash = qdisc_class_hash_alloc(nsize);
562 ohash = clhash->hash;
563 osize = clhash->hashsize;
566 for (i = 0; i < osize; i++) {
567 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
568 h = qdisc_class_hash(cl->classid, nmask);
569 hlist_add_head(&cl->hnode, &nhash[h]);
572 clhash->hash = nhash;
573 clhash->hashsize = nsize;
574 clhash->hashmask = nmask;
575 sch_tree_unlock(sch);
577 qdisc_class_hash_free(ohash, osize);
579 EXPORT_SYMBOL(qdisc_class_hash_grow);
581 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
583 unsigned int size = 4;
585 clhash->hash = qdisc_class_hash_alloc(size);
586 if (clhash->hash == NULL)
588 clhash->hashsize = size;
589 clhash->hashmask = size - 1;
590 clhash->hashelems = 0;
593 EXPORT_SYMBOL(qdisc_class_hash_init);
595 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
597 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
599 EXPORT_SYMBOL(qdisc_class_hash_destroy);
601 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
602 struct Qdisc_class_common *cl)
606 INIT_HLIST_NODE(&cl->hnode);
607 h = qdisc_class_hash(cl->classid, clhash->hashmask);
608 hlist_add_head(&cl->hnode, &clhash->hash[h]);
611 EXPORT_SYMBOL(qdisc_class_hash_insert);
613 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
614 struct Qdisc_class_common *cl)
616 hlist_del(&cl->hnode);
619 EXPORT_SYMBOL(qdisc_class_hash_remove);
621 /* Allocate an unique handle from space managed by kernel
622 * Possible range is [8000-FFFF]:0000 (0x8000 values)
624 static u32 qdisc_alloc_handle(struct net_device *dev)
627 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
630 autohandle += TC_H_MAKE(0x10000U, 0);
631 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
632 autohandle = TC_H_MAKE(0x80000000U, 0);
633 if (!qdisc_lookup(dev, autohandle))
641 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
643 const struct Qdisc_class_ops *cops;
649 while ((parentid = sch->parent)) {
650 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
653 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
655 WARN_ON(parentid != TC_H_ROOT);
658 cops = sch->ops->cl_ops;
659 if (cops->qlen_notify) {
660 cl = cops->get(sch, parentid);
661 cops->qlen_notify(sch, cl);
667 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
669 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
670 struct nlmsghdr *n, u32 clid,
671 struct Qdisc *old, struct Qdisc *new)
674 qdisc_notify(net, skb, n, clid, old, new);
680 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
683 * When appropriate send a netlink notification using 'skb'
686 * On success, destroy old qdisc.
689 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
690 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
691 struct Qdisc *new, struct Qdisc *old)
693 struct Qdisc *q = old;
694 struct net *net = dev_net(dev);
697 if (parent == NULL) {
698 unsigned int i, num_q, ingress;
701 num_q = dev->num_tx_queues;
702 if ((q && q->flags & TCQ_F_INGRESS) ||
703 (new && new->flags & TCQ_F_INGRESS)) {
706 if (!dev_ingress_queue(dev))
710 if (dev->flags & IFF_UP)
713 if (new && new->ops->attach) {
714 new->ops->attach(new);
718 for (i = 0; i < num_q; i++) {
719 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
722 dev_queue = netdev_get_tx_queue(dev, i);
724 old = dev_graft_qdisc(dev_queue, new);
726 atomic_inc(&new->refcnt);
733 notify_and_destroy(net, skb, n, classid,
735 if (new && !new->ops->attach)
736 atomic_inc(&new->refcnt);
737 dev->qdisc = new ? : &noop_qdisc;
739 notify_and_destroy(net, skb, n, classid, old, new);
742 if (dev->flags & IFF_UP)
745 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
748 if (cops && cops->graft) {
749 unsigned long cl = cops->get(parent, classid);
751 err = cops->graft(parent, cl, new, &old);
752 cops->put(parent, cl);
757 notify_and_destroy(net, skb, n, classid, old, new);
762 /* lockdep annotation is needed for ingress; egress gets it only for name */
763 static struct lock_class_key qdisc_tx_lock;
764 static struct lock_class_key qdisc_rx_lock;
767 Allocate and initialize new qdisc.
769 Parameters are passed via opt.
772 static struct Qdisc *
773 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
774 struct Qdisc *p, u32 parent, u32 handle,
775 struct nlattr **tca, int *errp)
778 struct nlattr *kind = tca[TCA_KIND];
780 struct Qdisc_ops *ops;
781 struct qdisc_size_table *stab;
783 ops = qdisc_lookup_ops(kind);
784 #ifdef CONFIG_MODULES
785 if (ops == NULL && kind != NULL) {
787 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
788 /* We dropped the RTNL semaphore in order to
789 * perform the module load. So, even if we
790 * succeeded in loading the module we have to
791 * tell the caller to replay the request. We
792 * indicate this using -EAGAIN.
793 * We replay the request because the device may
794 * go away in the mean time.
797 request_module("sch_%s", name);
799 ops = qdisc_lookup_ops(kind);
801 /* We will try again qdisc_lookup_ops,
802 * so don't keep a reference.
804 module_put(ops->owner);
816 sch = qdisc_alloc(dev_queue, ops);
822 sch->parent = parent;
824 if (handle == TC_H_INGRESS) {
825 sch->flags |= TCQ_F_INGRESS;
826 handle = TC_H_MAKE(TC_H_INGRESS, 0);
827 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
830 handle = qdisc_alloc_handle(dev);
835 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
836 if (!netif_is_multiqueue(dev))
837 sch->flags |= TCQ_F_ONETXQUEUE;
840 sch->handle = handle;
842 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
844 stab = qdisc_get_stab(tca[TCA_STAB]);
849 rcu_assign_pointer(sch->stab, stab);
852 spinlock_t *root_lock;
855 if (sch->flags & TCQ_F_MQROOT)
858 if ((sch->parent != TC_H_ROOT) &&
859 !(sch->flags & TCQ_F_INGRESS) &&
860 (!p || !(p->flags & TCQ_F_MQROOT)))
861 root_lock = qdisc_root_sleeping_lock(sch);
863 root_lock = qdisc_lock(sch);
865 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
866 root_lock, tca[TCA_RATE]);
877 kfree((char *) sch - sch->padded);
879 module_put(ops->owner);
886 * Any broken qdiscs that would require a ops->reset() here?
887 * The qdisc was never in action so it shouldn't be necessary.
889 qdisc_put_stab(rtnl_dereference(sch->stab));
895 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
897 struct qdisc_size_table *ostab, *stab = NULL;
900 if (tca[TCA_OPTIONS]) {
901 if (sch->ops->change == NULL)
903 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
909 stab = qdisc_get_stab(tca[TCA_STAB]);
911 return PTR_ERR(stab);
914 ostab = rtnl_dereference(sch->stab);
915 rcu_assign_pointer(sch->stab, stab);
916 qdisc_put_stab(ostab);
919 /* NB: ignores errors from replace_estimator
920 because change can't be undone. */
921 if (sch->flags & TCQ_F_MQROOT)
923 gen_replace_estimator(&sch->bstats, &sch->rate_est,
924 qdisc_root_sleeping_lock(sch),
931 struct check_loop_arg {
932 struct qdisc_walker w;
937 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
939 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
941 struct check_loop_arg arg;
943 if (q->ops->cl_ops == NULL)
946 arg.w.stop = arg.w.skip = arg.w.count = 0;
947 arg.w.fn = check_loop_fn;
950 q->ops->cl_ops->walk(q, &arg.w);
951 return arg.w.stop ? -ELOOP : 0;
955 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
958 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
959 struct check_loop_arg *arg = (struct check_loop_arg *)w;
961 leaf = cops->leaf(q, cl);
963 if (leaf == arg->p || arg->depth > 7)
965 return check_loop(leaf, arg->p, arg->depth + 1);
974 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
976 struct net *net = sock_net(skb->sk);
977 struct tcmsg *tcm = nlmsg_data(n);
978 struct nlattr *tca[TCA_MAX + 1];
979 struct net_device *dev;
981 struct Qdisc *q = NULL;
982 struct Qdisc *p = NULL;
985 if ((n->nlmsg_type != RTM_GETQDISC) && !capable(CAP_NET_ADMIN))
988 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
992 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
996 clid = tcm->tcm_parent;
998 if (clid != TC_H_ROOT) {
999 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1000 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1003 q = qdisc_leaf(p, clid);
1004 } else if (dev_ingress_queue(dev)) {
1005 q = dev_ingress_queue(dev)->qdisc_sleeping;
1013 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1016 q = qdisc_lookup(dev, tcm->tcm_handle);
1021 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1024 if (n->nlmsg_type == RTM_DELQDISC) {
1029 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1033 qdisc_notify(net, skb, n, clid, NULL, q);
1039 * Create/change qdisc.
1042 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1044 struct net *net = sock_net(skb->sk);
1046 struct nlattr *tca[TCA_MAX + 1];
1047 struct net_device *dev;
1049 struct Qdisc *q, *p;
1052 if (!capable(CAP_NET_ADMIN))
1056 /* Reinit, just in case something touches this. */
1057 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1061 tcm = nlmsg_data(n);
1062 clid = tcm->tcm_parent;
1065 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1071 if (clid != TC_H_ROOT) {
1072 if (clid != TC_H_INGRESS) {
1073 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1076 q = qdisc_leaf(p, clid);
1077 } else if (dev_ingress_queue_create(dev)) {
1078 q = dev_ingress_queue(dev)->qdisc_sleeping;
1084 /* It may be default qdisc, ignore it */
1085 if (q && q->handle == 0)
1088 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1089 if (tcm->tcm_handle) {
1090 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1092 if (TC_H_MIN(tcm->tcm_handle))
1094 q = qdisc_lookup(dev, tcm->tcm_handle);
1096 goto create_n_graft;
1097 if (n->nlmsg_flags & NLM_F_EXCL)
1099 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1102 (p && check_loop(q, p, 0)))
1104 atomic_inc(&q->refcnt);
1108 goto create_n_graft;
1110 /* This magic test requires explanation.
1112 * We know, that some child q is already
1113 * attached to this parent and have choice:
1114 * either to change it or to create/graft new one.
1116 * 1. We are allowed to create/graft only
1117 * if CREATE and REPLACE flags are set.
1119 * 2. If EXCL is set, requestor wanted to say,
1120 * that qdisc tcm_handle is not expected
1121 * to exist, so that we choose create/graft too.
1123 * 3. The last case is when no flags are set.
1124 * Alas, it is sort of hole in API, we
1125 * cannot decide what to do unambiguously.
1126 * For now we select create/graft, if
1127 * user gave KIND, which does not match existing.
1129 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1130 (n->nlmsg_flags & NLM_F_REPLACE) &&
1131 ((n->nlmsg_flags & NLM_F_EXCL) ||
1133 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1134 goto create_n_graft;
1138 if (!tcm->tcm_handle)
1140 q = qdisc_lookup(dev, tcm->tcm_handle);
1143 /* Change qdisc parameters */
1146 if (n->nlmsg_flags & NLM_F_EXCL)
1148 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1150 err = qdisc_change(q, tca);
1152 qdisc_notify(net, skb, n, clid, NULL, q);
1156 if (!(n->nlmsg_flags & NLM_F_CREATE))
1158 if (clid == TC_H_INGRESS) {
1159 if (dev_ingress_queue(dev))
1160 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1161 tcm->tcm_parent, tcm->tcm_parent,
1166 struct netdev_queue *dev_queue;
1168 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1169 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1171 dev_queue = p->dev_queue;
1173 dev_queue = netdev_get_tx_queue(dev, 0);
1175 q = qdisc_create(dev, dev_queue, p,
1176 tcm->tcm_parent, tcm->tcm_handle,
1186 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1196 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1197 u32 portid, u32 seq, u16 flags, int event)
1200 struct nlmsghdr *nlh;
1201 unsigned char *b = skb_tail_pointer(skb);
1203 struct qdisc_size_table *stab;
1205 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1207 goto out_nlmsg_trim;
1208 tcm = nlmsg_data(nlh);
1209 tcm->tcm_family = AF_UNSPEC;
1212 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1213 tcm->tcm_parent = clid;
1214 tcm->tcm_handle = q->handle;
1215 tcm->tcm_info = atomic_read(&q->refcnt);
1216 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1217 goto nla_put_failure;
1218 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1219 goto nla_put_failure;
1220 q->qstats.qlen = q->q.qlen;
1222 stab = rtnl_dereference(q->stab);
1223 if (stab && qdisc_dump_stab(skb, stab) < 0)
1224 goto nla_put_failure;
1226 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1227 qdisc_root_sleeping_lock(q), &d) < 0)
1228 goto nla_put_failure;
1230 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1231 goto nla_put_failure;
1233 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1234 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1235 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1236 goto nla_put_failure;
1238 if (gnet_stats_finish_copy(&d) < 0)
1239 goto nla_put_failure;
1241 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1250 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1252 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1255 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1256 struct nlmsghdr *n, u32 clid,
1257 struct Qdisc *old, struct Qdisc *new)
1259 struct sk_buff *skb;
1260 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1262 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1266 if (old && !tc_qdisc_dump_ignore(old)) {
1267 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1268 0, RTM_DELQDISC) < 0)
1271 if (new && !tc_qdisc_dump_ignore(new)) {
1272 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1273 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1278 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1279 n->nlmsg_flags & NLM_F_ECHO);
1286 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1287 struct netlink_callback *cb,
1288 int *q_idx_p, int s_q_idx)
1290 int ret = 0, q_idx = *q_idx_p;
1297 if (q_idx < s_q_idx) {
1300 if (!tc_qdisc_dump_ignore(q) &&
1301 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1302 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1306 list_for_each_entry(q, &root->list, list) {
1307 if (q_idx < s_q_idx) {
1311 if (!tc_qdisc_dump_ignore(q) &&
1312 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1313 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1326 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1328 struct net *net = sock_net(skb->sk);
1331 struct net_device *dev;
1333 s_idx = cb->args[0];
1334 s_q_idx = q_idx = cb->args[1];
1338 for_each_netdev_rcu(net, dev) {
1339 struct netdev_queue *dev_queue;
1347 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1350 dev_queue = dev_ingress_queue(dev);
1352 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1353 &q_idx, s_q_idx) < 0)
1364 cb->args[1] = q_idx;
1371 /************************************************
1372 * Traffic classes manipulation. *
1373 ************************************************/
1377 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1379 struct net *net = sock_net(skb->sk);
1380 struct tcmsg *tcm = nlmsg_data(n);
1381 struct nlattr *tca[TCA_MAX + 1];
1382 struct net_device *dev;
1383 struct Qdisc *q = NULL;
1384 const struct Qdisc_class_ops *cops;
1385 unsigned long cl = 0;
1386 unsigned long new_cl;
1392 if ((n->nlmsg_type != RTM_GETTCLASS) && !capable(CAP_NET_ADMIN))
1395 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1399 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1404 parent == TC_H_UNSPEC - unspecified parent.
1405 parent == TC_H_ROOT - class is root, which has no parent.
1406 parent == X:0 - parent is root class.
1407 parent == X:Y - parent is a node in hierarchy.
1408 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1410 handle == 0:0 - generate handle from kernel pool.
1411 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1412 handle == X:Y - clear.
1413 handle == X:0 - root class.
1416 /* Step 1. Determine qdisc handle X:0 */
1418 portid = tcm->tcm_parent;
1419 clid = tcm->tcm_handle;
1420 qid = TC_H_MAJ(clid);
1422 if (portid != TC_H_ROOT) {
1423 u32 qid1 = TC_H_MAJ(portid);
1426 /* If both majors are known, they must be identical. */
1431 } else if (qid == 0)
1432 qid = dev->qdisc->handle;
1434 /* Now qid is genuine qdisc handle consistent
1435 * both with parent and child.
1437 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1440 portid = TC_H_MAKE(qid, portid);
1443 qid = dev->qdisc->handle;
1446 /* OK. Locate qdisc */
1447 q = qdisc_lookup(dev, qid);
1451 /* An check that it supports classes */
1452 cops = q->ops->cl_ops;
1456 /* Now try to get class */
1458 if (portid == TC_H_ROOT)
1461 clid = TC_H_MAKE(qid, clid);
1464 cl = cops->get(q, clid);
1468 if (n->nlmsg_type != RTM_NEWTCLASS ||
1469 !(n->nlmsg_flags & NLM_F_CREATE))
1472 switch (n->nlmsg_type) {
1475 if (n->nlmsg_flags & NLM_F_EXCL)
1481 err = cops->delete(q, cl);
1483 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1486 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1497 err = cops->change(q, clid, portid, tca, &new_cl);
1499 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1509 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1511 u32 portid, u32 seq, u16 flags, int event)
1514 struct nlmsghdr *nlh;
1515 unsigned char *b = skb_tail_pointer(skb);
1517 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1519 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1521 goto out_nlmsg_trim;
1522 tcm = nlmsg_data(nlh);
1523 tcm->tcm_family = AF_UNSPEC;
1526 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1527 tcm->tcm_parent = q->handle;
1528 tcm->tcm_handle = q->handle;
1530 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1531 goto nla_put_failure;
1532 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1533 goto nla_put_failure;
1535 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1536 qdisc_root_sleeping_lock(q), &d) < 0)
1537 goto nla_put_failure;
1539 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1540 goto nla_put_failure;
1542 if (gnet_stats_finish_copy(&d) < 0)
1543 goto nla_put_failure;
1545 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1554 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1555 struct nlmsghdr *n, struct Qdisc *q,
1556 unsigned long cl, int event)
1558 struct sk_buff *skb;
1559 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1561 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1565 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1570 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1571 n->nlmsg_flags & NLM_F_ECHO);
1574 struct qdisc_dump_args {
1575 struct qdisc_walker w;
1576 struct sk_buff *skb;
1577 struct netlink_callback *cb;
1580 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1582 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1584 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1585 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1588 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1589 struct tcmsg *tcm, struct netlink_callback *cb,
1592 struct qdisc_dump_args arg;
1594 if (tc_qdisc_dump_ignore(q) ||
1595 *t_p < s_t || !q->ops->cl_ops ||
1597 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1602 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1603 arg.w.fn = qdisc_class_dump;
1607 arg.w.skip = cb->args[1];
1609 q->ops->cl_ops->walk(q, &arg.w);
1610 cb->args[1] = arg.w.count;
1617 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1618 struct tcmsg *tcm, struct netlink_callback *cb,
1626 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1629 list_for_each_entry(q, &root->list, list) {
1630 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1637 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1639 struct tcmsg *tcm = nlmsg_data(cb->nlh);
1640 struct net *net = sock_net(skb->sk);
1641 struct netdev_queue *dev_queue;
1642 struct net_device *dev;
1645 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1647 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1654 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1657 dev_queue = dev_ingress_queue(dev);
1659 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1670 /* Main classifier routine: scans classifier chain attached
1671 * to this qdisc, (optionally) tests for protocol and asks
1672 * specific classifiers.
1674 int tc_classify_compat(struct sk_buff *skb, const struct tcf_proto *tp,
1675 struct tcf_result *res)
1677 __be16 protocol = skb->protocol;
1680 for (; tp; tp = tp->next) {
1681 if (tp->protocol != protocol &&
1682 tp->protocol != htons(ETH_P_ALL))
1684 err = tp->classify(skb, tp, res);
1687 #ifdef CONFIG_NET_CLS_ACT
1688 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1689 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1696 EXPORT_SYMBOL(tc_classify_compat);
1698 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1699 struct tcf_result *res)
1702 #ifdef CONFIG_NET_CLS_ACT
1703 const struct tcf_proto *otp = tp;
1707 err = tc_classify_compat(skb, tp, res);
1708 #ifdef CONFIG_NET_CLS_ACT
1709 if (err == TC_ACT_RECLASSIFY) {
1710 u32 verd = G_TC_VERD(skb->tc_verd);
1713 if (verd++ >= MAX_REC_LOOP) {
1714 net_notice_ratelimited("%s: packet reclassify loop rule prio %u protocol %02x\n",
1717 ntohs(tp->protocol));
1720 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1726 EXPORT_SYMBOL(tc_classify);
1728 void tcf_destroy(struct tcf_proto *tp)
1730 tp->ops->destroy(tp);
1731 module_put(tp->ops->owner);
1735 void tcf_destroy_chain(struct tcf_proto **fl)
1737 struct tcf_proto *tp;
1739 while ((tp = *fl) != NULL) {
1744 EXPORT_SYMBOL(tcf_destroy_chain);
1746 #ifdef CONFIG_PROC_FS
1747 static int psched_show(struct seq_file *seq, void *v)
1751 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1752 seq_printf(seq, "%08x %08x %08x %08x\n",
1753 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1755 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1760 static int psched_open(struct inode *inode, struct file *file)
1762 return single_open(file, psched_show, NULL);
1765 static const struct file_operations psched_fops = {
1766 .owner = THIS_MODULE,
1767 .open = psched_open,
1769 .llseek = seq_lseek,
1770 .release = single_release,
1773 static int __net_init psched_net_init(struct net *net)
1775 struct proc_dir_entry *e;
1777 e = proc_create("psched", 0, net->proc_net, &psched_fops);
1784 static void __net_exit psched_net_exit(struct net *net)
1786 remove_proc_entry("psched", net->proc_net);
1789 static int __net_init psched_net_init(struct net *net)
1794 static void __net_exit psched_net_exit(struct net *net)
1799 static struct pernet_operations psched_net_ops = {
1800 .init = psched_net_init,
1801 .exit = psched_net_exit,
1804 static int __init pktsched_init(void)
1808 err = register_pernet_subsys(&psched_net_ops);
1810 pr_err("pktsched_init: "
1811 "cannot initialize per netns operations\n");
1815 register_qdisc(&pfifo_qdisc_ops);
1816 register_qdisc(&bfifo_qdisc_ops);
1817 register_qdisc(&pfifo_head_drop_qdisc_ops);
1818 register_qdisc(&mq_qdisc_ops);
1820 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1821 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1822 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1823 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1824 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1825 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1830 subsys_initcall(pktsched_init);