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>
32 #include <net/net_namespace.h>
34 #include <net/netlink.h>
35 #include <net/pkt_sched.h>
37 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid,
38 struct Qdisc *old, struct Qdisc *new);
39 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
40 struct Qdisc *q, unsigned long cl, int event);
47 This file consists of two interrelated parts:
49 1. queueing disciplines manager frontend.
50 2. traffic classes manager frontend.
52 Generally, queueing discipline ("qdisc") is a black box,
53 which is able to enqueue packets and to dequeue them (when
54 device is ready to send something) in order and at times
55 determined by algorithm hidden in it.
57 qdisc's are divided to two categories:
58 - "queues", which have no internal structure visible from outside.
59 - "schedulers", which split all the packets to "traffic classes",
60 using "packet classifiers" (look at cls_api.c)
62 In turn, classes may have child qdiscs (as rule, queues)
63 attached to them etc. etc. etc.
65 The goal of the routines in this file is to translate
66 information supplied by user in the form of handles
67 to more intelligible for kernel form, to make some sanity
68 checks and part of work, which is common to all qdiscs
69 and to provide rtnetlink notifications.
71 All real intelligent work is done inside qdisc modules.
75 Every discipline has two major routines: enqueue and dequeue.
79 dequeue usually returns a skb to send. It is allowed to return NULL,
80 but it does not mean that queue is empty, it just means that
81 discipline does not want to send anything this time.
82 Queue is really empty if q->q.qlen == 0.
83 For complicated disciplines with multiple queues q->q is not
84 real packet queue, but however q->q.qlen must be valid.
88 enqueue returns 0, if packet was enqueued successfully.
89 If packet (this one or another one) was dropped, it returns
91 NET_XMIT_DROP - this packet dropped
92 Expected action: do not backoff, but wait until queue will clear.
93 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
94 Expected action: backoff or ignore
95 NET_XMIT_POLICED - dropped by police.
96 Expected action: backoff or error to real-time apps.
102 like dequeue but without removing a packet from the queue
106 returns qdisc to initial state: purge all buffers, clear all
107 timers, counters (except for statistics) etc.
111 initializes newly created qdisc.
115 destroys resources allocated by init and during lifetime of qdisc.
119 changes qdisc parameters.
122 /* Protects list of registered TC modules. It is pure SMP lock. */
123 static DEFINE_RWLOCK(qdisc_mod_lock);
126 /************************************************
127 * Queueing disciplines manipulation. *
128 ************************************************/
131 /* The list of all installed queueing disciplines. */
133 static struct Qdisc_ops *qdisc_base;
135 /* Register/uregister queueing discipline */
137 int register_qdisc(struct Qdisc_ops *qops)
139 struct Qdisc_ops *q, **qp;
142 write_lock(&qdisc_mod_lock);
143 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
144 if (!strcmp(qops->id, q->id))
147 if (qops->enqueue == NULL)
148 qops->enqueue = noop_qdisc_ops.enqueue;
149 if (qops->peek == NULL) {
150 if (qops->dequeue == NULL) {
151 qops->peek = noop_qdisc_ops.peek;
157 if (qops->dequeue == NULL)
158 qops->dequeue = noop_qdisc_ops.dequeue;
164 write_unlock(&qdisc_mod_lock);
167 EXPORT_SYMBOL(register_qdisc);
169 int unregister_qdisc(struct Qdisc_ops *qops)
171 struct Qdisc_ops *q, **qp;
174 write_lock(&qdisc_mod_lock);
175 for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next)
183 write_unlock(&qdisc_mod_lock);
186 EXPORT_SYMBOL(unregister_qdisc);
188 /* We know handle. Find qdisc among all qdisc's attached to device
189 (root qdisc, all its children, children of children etc.)
192 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
196 if (!(root->flags & TCQ_F_BUILTIN) &&
197 root->handle == handle)
200 list_for_each_entry(q, &root->list, list) {
201 if (q->handle == handle)
207 static void qdisc_list_add(struct Qdisc *q)
209 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
210 list_add_tail(&q->list, &qdisc_dev(q)->qdisc->list);
213 void qdisc_list_del(struct Qdisc *q)
215 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
218 EXPORT_SYMBOL(qdisc_list_del);
220 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
224 q = qdisc_match_from_root(dev->qdisc, handle);
228 q = qdisc_match_from_root(dev->rx_queue.qdisc_sleeping, handle);
233 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
237 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
241 cl = cops->get(p, classid);
245 leaf = cops->leaf(p, cl);
250 /* Find queueing discipline by name */
252 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
254 struct Qdisc_ops *q = NULL;
257 read_lock(&qdisc_mod_lock);
258 for (q = qdisc_base; q; q = q->next) {
259 if (nla_strcmp(kind, q->id) == 0) {
260 if (!try_module_get(q->owner))
265 read_unlock(&qdisc_mod_lock);
270 static struct qdisc_rate_table *qdisc_rtab_list;
272 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
274 struct qdisc_rate_table *rtab;
276 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
277 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
283 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
284 nla_len(tab) != TC_RTAB_SIZE)
287 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
291 memcpy(rtab->data, nla_data(tab), 1024);
292 rtab->next = qdisc_rtab_list;
293 qdisc_rtab_list = rtab;
297 EXPORT_SYMBOL(qdisc_get_rtab);
299 void qdisc_put_rtab(struct qdisc_rate_table *tab)
301 struct qdisc_rate_table *rtab, **rtabp;
303 if (!tab || --tab->refcnt)
306 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
314 EXPORT_SYMBOL(qdisc_put_rtab);
316 static LIST_HEAD(qdisc_stab_list);
317 static DEFINE_SPINLOCK(qdisc_stab_lock);
319 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
320 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
321 [TCA_STAB_DATA] = { .type = NLA_BINARY },
324 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
326 struct nlattr *tb[TCA_STAB_MAX + 1];
327 struct qdisc_size_table *stab;
328 struct tc_sizespec *s;
329 unsigned int tsize = 0;
333 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
336 if (!tb[TCA_STAB_BASE])
337 return ERR_PTR(-EINVAL);
339 s = nla_data(tb[TCA_STAB_BASE]);
342 if (!tb[TCA_STAB_DATA])
343 return ERR_PTR(-EINVAL);
344 tab = nla_data(tb[TCA_STAB_DATA]);
345 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
348 if (!s || tsize != s->tsize || (!tab && tsize > 0))
349 return ERR_PTR(-EINVAL);
351 spin_lock(&qdisc_stab_lock);
353 list_for_each_entry(stab, &qdisc_stab_list, list) {
354 if (memcmp(&stab->szopts, s, sizeof(*s)))
356 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
359 spin_unlock(&qdisc_stab_lock);
363 spin_unlock(&qdisc_stab_lock);
365 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
367 return ERR_PTR(-ENOMEM);
372 memcpy(stab->data, tab, tsize * sizeof(u16));
374 spin_lock(&qdisc_stab_lock);
375 list_add_tail(&stab->list, &qdisc_stab_list);
376 spin_unlock(&qdisc_stab_lock);
381 void qdisc_put_stab(struct qdisc_size_table *tab)
386 spin_lock(&qdisc_stab_lock);
388 if (--tab->refcnt == 0) {
389 list_del(&tab->list);
393 spin_unlock(&qdisc_stab_lock);
395 EXPORT_SYMBOL(qdisc_put_stab);
397 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
401 nest = nla_nest_start(skb, TCA_STAB);
403 goto nla_put_failure;
404 NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts);
405 nla_nest_end(skb, nest);
413 void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab)
417 pkt_len = skb->len + stab->szopts.overhead;
418 if (unlikely(!stab->szopts.tsize))
421 slot = pkt_len + stab->szopts.cell_align;
422 if (unlikely(slot < 0))
425 slot >>= stab->szopts.cell_log;
426 if (likely(slot < stab->szopts.tsize))
427 pkt_len = stab->data[slot];
429 pkt_len = stab->data[stab->szopts.tsize - 1] *
430 (slot / stab->szopts.tsize) +
431 stab->data[slot % stab->szopts.tsize];
433 pkt_len <<= stab->szopts.size_log;
435 if (unlikely(pkt_len < 1))
437 qdisc_skb_cb(skb)->pkt_len = pkt_len;
439 EXPORT_SYMBOL(qdisc_calculate_pkt_len);
441 void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
443 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
445 "%s: %s qdisc %X: is non-work-conserving?\n",
446 txt, qdisc->ops->id, qdisc->handle >> 16);
447 qdisc->flags |= TCQ_F_WARN_NONWC;
450 EXPORT_SYMBOL(qdisc_warn_nonwc);
452 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
454 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
457 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
458 __netif_schedule(qdisc_root(wd->qdisc));
460 return HRTIMER_NORESTART;
463 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
465 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
466 wd->timer.function = qdisc_watchdog;
469 EXPORT_SYMBOL(qdisc_watchdog_init);
471 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
475 if (test_bit(__QDISC_STATE_DEACTIVATED,
476 &qdisc_root_sleeping(wd->qdisc)->state))
479 wd->qdisc->flags |= TCQ_F_THROTTLED;
480 time = ktime_set(0, 0);
481 time = ktime_add_ns(time, PSCHED_TICKS2NS(expires));
482 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
484 EXPORT_SYMBOL(qdisc_watchdog_schedule);
486 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
488 hrtimer_cancel(&wd->timer);
489 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
491 EXPORT_SYMBOL(qdisc_watchdog_cancel);
493 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
495 unsigned int size = n * sizeof(struct hlist_head), i;
496 struct hlist_head *h;
498 if (size <= PAGE_SIZE)
499 h = kmalloc(size, GFP_KERNEL);
501 h = (struct hlist_head *)
502 __get_free_pages(GFP_KERNEL, get_order(size));
505 for (i = 0; i < n; i++)
506 INIT_HLIST_HEAD(&h[i]);
511 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
513 unsigned int size = n * sizeof(struct hlist_head);
515 if (size <= PAGE_SIZE)
518 free_pages((unsigned long)h, get_order(size));
521 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
523 struct Qdisc_class_common *cl;
524 struct hlist_node *n, *next;
525 struct hlist_head *nhash, *ohash;
526 unsigned int nsize, nmask, osize;
529 /* Rehash when load factor exceeds 0.75 */
530 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
532 nsize = clhash->hashsize * 2;
534 nhash = qdisc_class_hash_alloc(nsize);
538 ohash = clhash->hash;
539 osize = clhash->hashsize;
542 for (i = 0; i < osize; i++) {
543 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
544 h = qdisc_class_hash(cl->classid, nmask);
545 hlist_add_head(&cl->hnode, &nhash[h]);
548 clhash->hash = nhash;
549 clhash->hashsize = nsize;
550 clhash->hashmask = nmask;
551 sch_tree_unlock(sch);
553 qdisc_class_hash_free(ohash, osize);
555 EXPORT_SYMBOL(qdisc_class_hash_grow);
557 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
559 unsigned int size = 4;
561 clhash->hash = qdisc_class_hash_alloc(size);
562 if (clhash->hash == NULL)
564 clhash->hashsize = size;
565 clhash->hashmask = size - 1;
566 clhash->hashelems = 0;
569 EXPORT_SYMBOL(qdisc_class_hash_init);
571 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
573 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
575 EXPORT_SYMBOL(qdisc_class_hash_destroy);
577 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
578 struct Qdisc_class_common *cl)
582 INIT_HLIST_NODE(&cl->hnode);
583 h = qdisc_class_hash(cl->classid, clhash->hashmask);
584 hlist_add_head(&cl->hnode, &clhash->hash[h]);
587 EXPORT_SYMBOL(qdisc_class_hash_insert);
589 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
590 struct Qdisc_class_common *cl)
592 hlist_del(&cl->hnode);
595 EXPORT_SYMBOL(qdisc_class_hash_remove);
597 /* Allocate an unique handle from space managed by kernel */
599 static u32 qdisc_alloc_handle(struct net_device *dev)
602 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
605 autohandle += TC_H_MAKE(0x10000U, 0);
606 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
607 autohandle = TC_H_MAKE(0x80000000U, 0);
608 } while (qdisc_lookup(dev, autohandle) && --i > 0);
610 return i>0 ? autohandle : 0;
613 /* Attach toplevel qdisc to device queue. */
615 static struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
618 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
619 spinlock_t *root_lock;
621 root_lock = qdisc_lock(oqdisc);
622 spin_lock_bh(root_lock);
624 /* Prune old scheduler */
625 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
628 /* ... and graft new one */
631 dev_queue->qdisc_sleeping = qdisc;
632 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
634 spin_unlock_bh(root_lock);
639 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
641 const struct Qdisc_class_ops *cops;
647 while ((parentid = sch->parent)) {
648 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
651 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
653 WARN_ON(parentid != TC_H_ROOT);
656 cops = sch->ops->cl_ops;
657 if (cops->qlen_notify) {
658 cl = cops->get(sch, parentid);
659 cops->qlen_notify(sch, cl);
665 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
667 static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid,
668 struct Qdisc *old, struct Qdisc *new)
671 qdisc_notify(skb, n, clid, old, new);
677 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
680 * When appropriate send a netlink notification using 'skb'
683 * On success, destroy old qdisc.
686 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
687 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
688 struct Qdisc *new, struct Qdisc *old)
690 struct Qdisc *q = old;
693 if (parent == NULL) {
694 unsigned int i, num_q, ingress;
697 num_q = dev->num_tx_queues;
698 if ((q && q->flags & TCQ_F_INGRESS) ||
699 (new && new->flags & TCQ_F_INGRESS)) {
704 if (dev->flags & IFF_UP)
707 for (i = 0; i < num_q; i++) {
708 struct netdev_queue *dev_queue = &dev->rx_queue;
711 dev_queue = netdev_get_tx_queue(dev, i);
713 old = dev_graft_qdisc(dev_queue, new);
715 atomic_inc(&new->refcnt);
720 notify_and_destroy(skb, n, classid, dev->qdisc, new);
722 atomic_inc(&new->refcnt);
723 dev->qdisc = new ? : &noop_qdisc;
725 if (dev->flags & IFF_UP)
728 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
731 if (cops && cops->graft) {
732 unsigned long cl = cops->get(parent, classid);
734 err = cops->graft(parent, cl, new, &old);
735 cops->put(parent, cl);
740 notify_and_destroy(skb, n, classid, old, new);
745 /* lockdep annotation is needed for ingress; egress gets it only for name */
746 static struct lock_class_key qdisc_tx_lock;
747 static struct lock_class_key qdisc_rx_lock;
750 Allocate and initialize new qdisc.
752 Parameters are passed via opt.
755 static struct Qdisc *
756 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
757 u32 parent, u32 handle, struct nlattr **tca, int *errp)
760 struct nlattr *kind = tca[TCA_KIND];
762 struct Qdisc_ops *ops;
763 struct qdisc_size_table *stab;
765 ops = qdisc_lookup_ops(kind);
766 #ifdef CONFIG_MODULES
767 if (ops == NULL && kind != NULL) {
769 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
770 /* We dropped the RTNL semaphore in order to
771 * perform the module load. So, even if we
772 * succeeded in loading the module we have to
773 * tell the caller to replay the request. We
774 * indicate this using -EAGAIN.
775 * We replay the request because the device may
776 * go away in the mean time.
779 request_module("sch_%s", name);
781 ops = qdisc_lookup_ops(kind);
783 /* We will try again qdisc_lookup_ops,
784 * so don't keep a reference.
786 module_put(ops->owner);
798 sch = qdisc_alloc(dev_queue, ops);
804 sch->parent = parent;
806 if (handle == TC_H_INGRESS) {
807 sch->flags |= TCQ_F_INGRESS;
808 handle = TC_H_MAKE(TC_H_INGRESS, 0);
809 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
812 handle = qdisc_alloc_handle(dev);
817 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
820 sch->handle = handle;
822 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
824 stab = qdisc_get_stab(tca[TCA_STAB]);
832 spinlock_t *root_lock;
834 if ((sch->parent != TC_H_ROOT) &&
835 !(sch->flags & TCQ_F_INGRESS))
836 root_lock = qdisc_root_sleeping_lock(sch);
838 root_lock = qdisc_lock(sch);
840 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
841 root_lock, tca[TCA_RATE]);
844 * Any broken qdiscs that would require
845 * a ops->reset() here? The qdisc was never
846 * in action so it shouldn't be necessary.
859 qdisc_put_stab(sch->stab);
861 kfree((char *) sch - sch->padded);
863 module_put(ops->owner);
869 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
871 struct qdisc_size_table *stab = NULL;
874 if (tca[TCA_OPTIONS]) {
875 if (sch->ops->change == NULL)
877 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
883 stab = qdisc_get_stab(tca[TCA_STAB]);
885 return PTR_ERR(stab);
888 qdisc_put_stab(sch->stab);
892 /* NB: ignores errors from replace_estimator
893 because change can't be undone. */
894 gen_replace_estimator(&sch->bstats, &sch->rate_est,
895 qdisc_root_sleeping_lock(sch),
901 struct check_loop_arg
903 struct qdisc_walker w;
908 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
910 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
912 struct check_loop_arg arg;
914 if (q->ops->cl_ops == NULL)
917 arg.w.stop = arg.w.skip = arg.w.count = 0;
918 arg.w.fn = check_loop_fn;
921 q->ops->cl_ops->walk(q, &arg.w);
922 return arg.w.stop ? -ELOOP : 0;
926 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
929 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
930 struct check_loop_arg *arg = (struct check_loop_arg *)w;
932 leaf = cops->leaf(q, cl);
934 if (leaf == arg->p || arg->depth > 7)
936 return check_loop(leaf, arg->p, arg->depth + 1);
945 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
947 struct net *net = sock_net(skb->sk);
948 struct tcmsg *tcm = NLMSG_DATA(n);
949 struct nlattr *tca[TCA_MAX + 1];
950 struct net_device *dev;
951 u32 clid = tcm->tcm_parent;
952 struct Qdisc *q = NULL;
953 struct Qdisc *p = NULL;
956 if (net != &init_net)
959 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
962 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
967 if (clid != TC_H_ROOT) {
968 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
969 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
971 q = qdisc_leaf(p, clid);
972 } else { /* ingress */
973 q = dev->rx_queue.qdisc_sleeping;
981 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
984 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
988 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
991 if (n->nlmsg_type == RTM_DELQDISC) {
996 if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0)
999 qdisc_notify(skb, n, clid, NULL, q);
1005 Create/change qdisc.
1008 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1010 struct net *net = sock_net(skb->sk);
1012 struct nlattr *tca[TCA_MAX + 1];
1013 struct net_device *dev;
1015 struct Qdisc *q, *p;
1018 if (net != &init_net)
1022 /* Reinit, just in case something touches this. */
1023 tcm = NLMSG_DATA(n);
1024 clid = tcm->tcm_parent;
1027 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1030 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1035 if (clid != TC_H_ROOT) {
1036 if (clid != TC_H_INGRESS) {
1037 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
1039 q = qdisc_leaf(p, clid);
1040 } else { /*ingress */
1041 q = dev->rx_queue.qdisc_sleeping;
1047 /* It may be default qdisc, ignore it */
1048 if (q && q->handle == 0)
1051 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1052 if (tcm->tcm_handle) {
1053 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
1055 if (TC_H_MIN(tcm->tcm_handle))
1057 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
1058 goto create_n_graft;
1059 if (n->nlmsg_flags&NLM_F_EXCL)
1061 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1064 (p && check_loop(q, p, 0)))
1066 atomic_inc(&q->refcnt);
1070 goto create_n_graft;
1072 /* This magic test requires explanation.
1074 * We know, that some child q is already
1075 * attached to this parent and have choice:
1076 * either to change it or to create/graft new one.
1078 * 1. We are allowed to create/graft only
1079 * if CREATE and REPLACE flags are set.
1081 * 2. If EXCL is set, requestor wanted to say,
1082 * that qdisc tcm_handle is not expected
1083 * to exist, so that we choose create/graft too.
1085 * 3. The last case is when no flags are set.
1086 * Alas, it is sort of hole in API, we
1087 * cannot decide what to do unambiguously.
1088 * For now we select create/graft, if
1089 * user gave KIND, which does not match existing.
1091 if ((n->nlmsg_flags&NLM_F_CREATE) &&
1092 (n->nlmsg_flags&NLM_F_REPLACE) &&
1093 ((n->nlmsg_flags&NLM_F_EXCL) ||
1095 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1096 goto create_n_graft;
1100 if (!tcm->tcm_handle)
1102 q = qdisc_lookup(dev, tcm->tcm_handle);
1105 /* Change qdisc parameters */
1108 if (n->nlmsg_flags&NLM_F_EXCL)
1110 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1112 err = qdisc_change(q, tca);
1114 qdisc_notify(skb, n, clid, NULL, q);
1118 if (!(n->nlmsg_flags&NLM_F_CREATE))
1120 if (clid == TC_H_INGRESS)
1121 q = qdisc_create(dev, &dev->rx_queue,
1122 tcm->tcm_parent, tcm->tcm_parent,
1125 q = qdisc_create(dev, netdev_get_tx_queue(dev, 0),
1126 tcm->tcm_parent, tcm->tcm_handle,
1135 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1145 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1146 u32 pid, u32 seq, u16 flags, int event)
1149 struct nlmsghdr *nlh;
1150 unsigned char *b = skb_tail_pointer(skb);
1153 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1154 tcm = NLMSG_DATA(nlh);
1155 tcm->tcm_family = AF_UNSPEC;
1158 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1159 tcm->tcm_parent = clid;
1160 tcm->tcm_handle = q->handle;
1161 tcm->tcm_info = atomic_read(&q->refcnt);
1162 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1163 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1164 goto nla_put_failure;
1165 q->qstats.qlen = q->q.qlen;
1167 if (q->stab && qdisc_dump_stab(skb, q->stab) < 0)
1168 goto nla_put_failure;
1170 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1171 qdisc_root_sleeping_lock(q), &d) < 0)
1172 goto nla_put_failure;
1174 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1175 goto nla_put_failure;
1177 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1178 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
1179 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1180 goto nla_put_failure;
1182 if (gnet_stats_finish_copy(&d) < 0)
1183 goto nla_put_failure;
1185 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1194 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1195 u32 clid, struct Qdisc *old, struct Qdisc *new)
1197 struct sk_buff *skb;
1198 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1200 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1204 if (old && old->handle) {
1205 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
1209 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1214 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1221 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1223 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1226 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1227 struct netlink_callback *cb,
1228 int *q_idx_p, int s_q_idx)
1230 int ret = 0, q_idx = *q_idx_p;
1237 if (q_idx < s_q_idx) {
1240 if (!tc_qdisc_dump_ignore(q) &&
1241 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1242 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1246 list_for_each_entry(q, &root->list, list) {
1247 if (q_idx < s_q_idx) {
1251 if (!tc_qdisc_dump_ignore(q) &&
1252 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1253 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1266 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1268 struct net *net = sock_net(skb->sk);
1271 struct net_device *dev;
1273 if (net != &init_net)
1276 s_idx = cb->args[0];
1277 s_q_idx = q_idx = cb->args[1];
1278 read_lock(&dev_base_lock);
1280 for_each_netdev(&init_net, dev) {
1281 struct netdev_queue *dev_queue;
1289 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1292 dev_queue = &dev->rx_queue;
1293 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1301 read_unlock(&dev_base_lock);
1304 cb->args[1] = q_idx;
1311 /************************************************
1312 * Traffic classes manipulation. *
1313 ************************************************/
1317 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1319 struct net *net = sock_net(skb->sk);
1320 struct tcmsg *tcm = NLMSG_DATA(n);
1321 struct nlattr *tca[TCA_MAX + 1];
1322 struct net_device *dev;
1323 struct Qdisc *q = NULL;
1324 const struct Qdisc_class_ops *cops;
1325 unsigned long cl = 0;
1326 unsigned long new_cl;
1327 u32 pid = tcm->tcm_parent;
1328 u32 clid = tcm->tcm_handle;
1329 u32 qid = TC_H_MAJ(clid);
1332 if (net != &init_net)
1335 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1338 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1343 parent == TC_H_UNSPEC - unspecified parent.
1344 parent == TC_H_ROOT - class is root, which has no parent.
1345 parent == X:0 - parent is root class.
1346 parent == X:Y - parent is a node in hierarchy.
1347 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1349 handle == 0:0 - generate handle from kernel pool.
1350 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1351 handle == X:Y - clear.
1352 handle == X:0 - root class.
1355 /* Step 1. Determine qdisc handle X:0 */
1357 if (pid != TC_H_ROOT) {
1358 u32 qid1 = TC_H_MAJ(pid);
1361 /* If both majors are known, they must be identical. */
1366 } else if (qid == 0)
1367 qid = dev->qdisc->handle;
1369 /* Now qid is genuine qdisc handle consistent
1370 both with parent and child.
1372 TC_H_MAJ(pid) still may be unspecified, complete it now.
1375 pid = TC_H_MAKE(qid, pid);
1378 qid = dev->qdisc->handle;
1381 /* OK. Locate qdisc */
1382 if ((q = qdisc_lookup(dev, qid)) == NULL)
1385 /* An check that it supports classes */
1386 cops = q->ops->cl_ops;
1390 /* Now try to get class */
1392 if (pid == TC_H_ROOT)
1395 clid = TC_H_MAKE(qid, clid);
1398 cl = cops->get(q, clid);
1402 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1405 switch (n->nlmsg_type) {
1408 if (n->nlmsg_flags&NLM_F_EXCL)
1414 err = cops->delete(q, cl);
1416 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1419 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1430 err = cops->change(q, clid, pid, tca, &new_cl);
1432 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1442 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1444 u32 pid, u32 seq, u16 flags, int event)
1447 struct nlmsghdr *nlh;
1448 unsigned char *b = skb_tail_pointer(skb);
1450 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1452 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1453 tcm = NLMSG_DATA(nlh);
1454 tcm->tcm_family = AF_UNSPEC;
1455 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1456 tcm->tcm_parent = q->handle;
1457 tcm->tcm_handle = q->handle;
1459 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1460 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1461 goto nla_put_failure;
1463 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1464 qdisc_root_sleeping_lock(q), &d) < 0)
1465 goto nla_put_failure;
1467 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1468 goto nla_put_failure;
1470 if (gnet_stats_finish_copy(&d) < 0)
1471 goto nla_put_failure;
1473 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1482 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1483 struct Qdisc *q, unsigned long cl, int event)
1485 struct sk_buff *skb;
1486 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1488 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1492 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1497 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1500 struct qdisc_dump_args
1502 struct qdisc_walker w;
1503 struct sk_buff *skb;
1504 struct netlink_callback *cb;
1507 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1509 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1511 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1512 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1515 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1516 struct tcmsg *tcm, struct netlink_callback *cb,
1519 struct qdisc_dump_args arg;
1521 if (tc_qdisc_dump_ignore(q) ||
1522 *t_p < s_t || !q->ops->cl_ops ||
1524 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1529 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1530 arg.w.fn = qdisc_class_dump;
1534 arg.w.skip = cb->args[1];
1536 q->ops->cl_ops->walk(q, &arg.w);
1537 cb->args[1] = arg.w.count;
1544 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1545 struct tcmsg *tcm, struct netlink_callback *cb,
1553 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1556 list_for_each_entry(q, &root->list, list) {
1557 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1564 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1566 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1567 struct net *net = sock_net(skb->sk);
1568 struct netdev_queue *dev_queue;
1569 struct net_device *dev;
1572 if (net != &init_net)
1575 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1577 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1583 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1586 dev_queue = &dev->rx_queue;
1587 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1597 /* Main classifier routine: scans classifier chain attached
1598 to this qdisc, (optionally) tests for protocol and asks
1599 specific classifiers.
1601 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1602 struct tcf_result *res)
1604 __be16 protocol = skb->protocol;
1607 for (; tp; tp = tp->next) {
1608 if ((tp->protocol == protocol ||
1609 tp->protocol == htons(ETH_P_ALL)) &&
1610 (err = tp->classify(skb, tp, res)) >= 0) {
1611 #ifdef CONFIG_NET_CLS_ACT
1612 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1613 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1620 EXPORT_SYMBOL(tc_classify_compat);
1622 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1623 struct tcf_result *res)
1627 #ifdef CONFIG_NET_CLS_ACT
1628 struct tcf_proto *otp = tp;
1631 protocol = skb->protocol;
1633 err = tc_classify_compat(skb, tp, res);
1634 #ifdef CONFIG_NET_CLS_ACT
1635 if (err == TC_ACT_RECLASSIFY) {
1636 u32 verd = G_TC_VERD(skb->tc_verd);
1639 if (verd++ >= MAX_REC_LOOP) {
1640 printk("rule prio %u protocol %02x reclassify loop, "
1642 tp->prio&0xffff, ntohs(tp->protocol));
1645 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1651 EXPORT_SYMBOL(tc_classify);
1653 void tcf_destroy(struct tcf_proto *tp)
1655 tp->ops->destroy(tp);
1656 module_put(tp->ops->owner);
1660 void tcf_destroy_chain(struct tcf_proto **fl)
1662 struct tcf_proto *tp;
1664 while ((tp = *fl) != NULL) {
1669 EXPORT_SYMBOL(tcf_destroy_chain);
1671 #ifdef CONFIG_PROC_FS
1672 static int psched_show(struct seq_file *seq, void *v)
1676 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1677 seq_printf(seq, "%08x %08x %08x %08x\n",
1678 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1680 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1685 static int psched_open(struct inode *inode, struct file *file)
1687 return single_open(file, psched_show, PDE(inode)->data);
1690 static const struct file_operations psched_fops = {
1691 .owner = THIS_MODULE,
1692 .open = psched_open,
1694 .llseek = seq_lseek,
1695 .release = single_release,
1699 static int __init pktsched_init(void)
1701 register_qdisc(&pfifo_qdisc_ops);
1702 register_qdisc(&bfifo_qdisc_ops);
1703 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1705 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1706 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1707 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1708 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1709 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1710 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1715 subsys_initcall(pktsched_init);