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_root_sleeping(q)->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)
225 for (i = 0; i < dev->num_tx_queues; i++) {
226 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
227 struct Qdisc *txq_root = txq->qdisc_sleeping;
229 q = qdisc_match_from_root(txq_root, handle);
234 q = qdisc_match_from_root(dev->rx_queue.qdisc_sleeping, handle);
239 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
243 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
247 cl = cops->get(p, classid);
251 leaf = cops->leaf(p, cl);
256 /* Find queueing discipline by name */
258 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
260 struct Qdisc_ops *q = NULL;
263 read_lock(&qdisc_mod_lock);
264 for (q = qdisc_base; q; q = q->next) {
265 if (nla_strcmp(kind, q->id) == 0) {
266 if (!try_module_get(q->owner))
271 read_unlock(&qdisc_mod_lock);
276 static struct qdisc_rate_table *qdisc_rtab_list;
278 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
280 struct qdisc_rate_table *rtab;
282 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
283 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
289 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
290 nla_len(tab) != TC_RTAB_SIZE)
293 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
297 memcpy(rtab->data, nla_data(tab), 1024);
298 rtab->next = qdisc_rtab_list;
299 qdisc_rtab_list = rtab;
303 EXPORT_SYMBOL(qdisc_get_rtab);
305 void qdisc_put_rtab(struct qdisc_rate_table *tab)
307 struct qdisc_rate_table *rtab, **rtabp;
309 if (!tab || --tab->refcnt)
312 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
320 EXPORT_SYMBOL(qdisc_put_rtab);
322 static LIST_HEAD(qdisc_stab_list);
323 static DEFINE_SPINLOCK(qdisc_stab_lock);
325 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
326 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
327 [TCA_STAB_DATA] = { .type = NLA_BINARY },
330 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
332 struct nlattr *tb[TCA_STAB_MAX + 1];
333 struct qdisc_size_table *stab;
334 struct tc_sizespec *s;
335 unsigned int tsize = 0;
339 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
342 if (!tb[TCA_STAB_BASE])
343 return ERR_PTR(-EINVAL);
345 s = nla_data(tb[TCA_STAB_BASE]);
348 if (!tb[TCA_STAB_DATA])
349 return ERR_PTR(-EINVAL);
350 tab = nla_data(tb[TCA_STAB_DATA]);
351 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
354 if (!s || tsize != s->tsize || (!tab && tsize > 0))
355 return ERR_PTR(-EINVAL);
357 spin_lock(&qdisc_stab_lock);
359 list_for_each_entry(stab, &qdisc_stab_list, list) {
360 if (memcmp(&stab->szopts, s, sizeof(*s)))
362 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
365 spin_unlock(&qdisc_stab_lock);
369 spin_unlock(&qdisc_stab_lock);
371 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
373 return ERR_PTR(-ENOMEM);
378 memcpy(stab->data, tab, tsize * sizeof(u16));
380 spin_lock(&qdisc_stab_lock);
381 list_add_tail(&stab->list, &qdisc_stab_list);
382 spin_unlock(&qdisc_stab_lock);
387 void qdisc_put_stab(struct qdisc_size_table *tab)
392 spin_lock(&qdisc_stab_lock);
394 if (--tab->refcnt == 0) {
395 list_del(&tab->list);
399 spin_unlock(&qdisc_stab_lock);
401 EXPORT_SYMBOL(qdisc_put_stab);
403 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
407 nest = nla_nest_start(skb, TCA_STAB);
409 goto nla_put_failure;
410 NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts);
411 nla_nest_end(skb, nest);
419 void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab)
423 pkt_len = skb->len + stab->szopts.overhead;
424 if (unlikely(!stab->szopts.tsize))
427 slot = pkt_len + stab->szopts.cell_align;
428 if (unlikely(slot < 0))
431 slot >>= stab->szopts.cell_log;
432 if (likely(slot < stab->szopts.tsize))
433 pkt_len = stab->data[slot];
435 pkt_len = stab->data[stab->szopts.tsize - 1] *
436 (slot / stab->szopts.tsize) +
437 stab->data[slot % stab->szopts.tsize];
439 pkt_len <<= stab->szopts.size_log;
441 if (unlikely(pkt_len < 1))
443 qdisc_skb_cb(skb)->pkt_len = pkt_len;
445 EXPORT_SYMBOL(qdisc_calculate_pkt_len);
447 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
449 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
452 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
453 __netif_schedule(qdisc_root(wd->qdisc));
455 return HRTIMER_NORESTART;
458 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
460 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
461 wd->timer.function = qdisc_watchdog;
464 EXPORT_SYMBOL(qdisc_watchdog_init);
466 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
470 if (test_bit(__QDISC_STATE_DEACTIVATED,
471 &qdisc_root_sleeping(wd->qdisc)->state))
474 wd->qdisc->flags |= TCQ_F_THROTTLED;
475 time = ktime_set(0, 0);
476 time = ktime_add_ns(time, PSCHED_US2NS(expires));
477 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
479 EXPORT_SYMBOL(qdisc_watchdog_schedule);
481 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
483 hrtimer_cancel(&wd->timer);
484 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
486 EXPORT_SYMBOL(qdisc_watchdog_cancel);
488 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
490 unsigned int size = n * sizeof(struct hlist_head), i;
491 struct hlist_head *h;
493 if (size <= PAGE_SIZE)
494 h = kmalloc(size, GFP_KERNEL);
496 h = (struct hlist_head *)
497 __get_free_pages(GFP_KERNEL, get_order(size));
500 for (i = 0; i < n; i++)
501 INIT_HLIST_HEAD(&h[i]);
506 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
508 unsigned int size = n * sizeof(struct hlist_head);
510 if (size <= PAGE_SIZE)
513 free_pages((unsigned long)h, get_order(size));
516 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
518 struct Qdisc_class_common *cl;
519 struct hlist_node *n, *next;
520 struct hlist_head *nhash, *ohash;
521 unsigned int nsize, nmask, osize;
524 /* Rehash when load factor exceeds 0.75 */
525 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
527 nsize = clhash->hashsize * 2;
529 nhash = qdisc_class_hash_alloc(nsize);
533 ohash = clhash->hash;
534 osize = clhash->hashsize;
537 for (i = 0; i < osize; i++) {
538 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
539 h = qdisc_class_hash(cl->classid, nmask);
540 hlist_add_head(&cl->hnode, &nhash[h]);
543 clhash->hash = nhash;
544 clhash->hashsize = nsize;
545 clhash->hashmask = nmask;
546 sch_tree_unlock(sch);
548 qdisc_class_hash_free(ohash, osize);
550 EXPORT_SYMBOL(qdisc_class_hash_grow);
552 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
554 unsigned int size = 4;
556 clhash->hash = qdisc_class_hash_alloc(size);
557 if (clhash->hash == NULL)
559 clhash->hashsize = size;
560 clhash->hashmask = size - 1;
561 clhash->hashelems = 0;
564 EXPORT_SYMBOL(qdisc_class_hash_init);
566 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
568 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
570 EXPORT_SYMBOL(qdisc_class_hash_destroy);
572 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
573 struct Qdisc_class_common *cl)
577 INIT_HLIST_NODE(&cl->hnode);
578 h = qdisc_class_hash(cl->classid, clhash->hashmask);
579 hlist_add_head(&cl->hnode, &clhash->hash[h]);
582 EXPORT_SYMBOL(qdisc_class_hash_insert);
584 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
585 struct Qdisc_class_common *cl)
587 hlist_del(&cl->hnode);
590 EXPORT_SYMBOL(qdisc_class_hash_remove);
592 /* Allocate an unique handle from space managed by kernel */
594 static u32 qdisc_alloc_handle(struct net_device *dev)
597 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
600 autohandle += TC_H_MAKE(0x10000U, 0);
601 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
602 autohandle = TC_H_MAKE(0x80000000U, 0);
603 } while (qdisc_lookup(dev, autohandle) && --i > 0);
605 return i>0 ? autohandle : 0;
608 /* Attach toplevel qdisc to device queue. */
610 static struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
613 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
614 spinlock_t *root_lock;
616 root_lock = qdisc_lock(oqdisc);
617 spin_lock_bh(root_lock);
619 /* Prune old scheduler */
620 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
623 /* ... and graft new one */
626 dev_queue->qdisc_sleeping = qdisc;
627 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
629 spin_unlock_bh(root_lock);
634 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
636 const struct Qdisc_class_ops *cops;
642 while ((parentid = sch->parent)) {
643 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
646 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
648 WARN_ON(parentid != TC_H_ROOT);
651 cops = sch->ops->cl_ops;
652 if (cops->qlen_notify) {
653 cl = cops->get(sch, parentid);
654 cops->qlen_notify(sch, cl);
660 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
662 static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid,
663 struct Qdisc *old, struct Qdisc *new)
666 qdisc_notify(skb, n, clid, old, new);
672 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
675 * When appropriate send a netlink notification using 'skb'
678 * On success, destroy old qdisc.
681 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
682 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
683 struct Qdisc *new, struct Qdisc *old)
685 struct Qdisc *q = old;
688 if (parent == NULL) {
689 unsigned int i, num_q, ingress;
692 num_q = dev->num_tx_queues;
693 if ((q && q->flags & TCQ_F_INGRESS) ||
694 (new && new->flags & TCQ_F_INGRESS)) {
699 if (dev->flags & IFF_UP)
702 for (i = 0; i < num_q; i++) {
703 struct netdev_queue *dev_queue = &dev->rx_queue;
706 dev_queue = netdev_get_tx_queue(dev, i);
708 old = dev_graft_qdisc(dev_queue, new);
710 atomic_inc(&new->refcnt);
712 notify_and_destroy(skb, n, classid, old, new);
715 if (dev->flags & IFF_UP)
718 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
723 unsigned long cl = cops->get(parent, classid);
725 err = cops->graft(parent, cl, new, &old);
726 cops->put(parent, cl);
730 notify_and_destroy(skb, n, classid, old, new);
735 /* lockdep annotation is needed for ingress; egress gets it only for name */
736 static struct lock_class_key qdisc_tx_lock;
737 static struct lock_class_key qdisc_rx_lock;
740 Allocate and initialize new qdisc.
742 Parameters are passed via opt.
745 static struct Qdisc *
746 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
747 u32 parent, u32 handle, struct nlattr **tca, int *errp)
750 struct nlattr *kind = tca[TCA_KIND];
752 struct Qdisc_ops *ops;
753 struct qdisc_size_table *stab;
755 ops = qdisc_lookup_ops(kind);
756 #ifdef CONFIG_MODULES
757 if (ops == NULL && kind != NULL) {
759 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
760 /* We dropped the RTNL semaphore in order to
761 * perform the module load. So, even if we
762 * succeeded in loading the module we have to
763 * tell the caller to replay the request. We
764 * indicate this using -EAGAIN.
765 * We replay the request because the device may
766 * go away in the mean time.
769 request_module("sch_%s", name);
771 ops = qdisc_lookup_ops(kind);
773 /* We will try again qdisc_lookup_ops,
774 * so don't keep a reference.
776 module_put(ops->owner);
788 sch = qdisc_alloc(dev_queue, ops);
794 sch->parent = parent;
796 if (handle == TC_H_INGRESS) {
797 sch->flags |= TCQ_F_INGRESS;
798 handle = TC_H_MAKE(TC_H_INGRESS, 0);
799 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
802 handle = qdisc_alloc_handle(dev);
807 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
810 sch->handle = handle;
812 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
814 stab = qdisc_get_stab(tca[TCA_STAB]);
822 spinlock_t *root_lock;
824 if ((sch->parent != TC_H_ROOT) &&
825 !(sch->flags & TCQ_F_INGRESS))
826 root_lock = qdisc_root_sleeping_lock(sch);
828 root_lock = qdisc_lock(sch);
830 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
831 root_lock, tca[TCA_RATE]);
834 * Any broken qdiscs that would require
835 * a ops->reset() here? The qdisc was never
836 * in action so it shouldn't be necessary.
849 qdisc_put_stab(sch->stab);
851 kfree((char *) sch - sch->padded);
853 module_put(ops->owner);
859 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
861 struct qdisc_size_table *stab = NULL;
864 if (tca[TCA_OPTIONS]) {
865 if (sch->ops->change == NULL)
867 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
873 stab = qdisc_get_stab(tca[TCA_STAB]);
875 return PTR_ERR(stab);
878 qdisc_put_stab(sch->stab);
882 /* NB: ignores errors from replace_estimator
883 because change can't be undone. */
884 gen_replace_estimator(&sch->bstats, &sch->rate_est,
885 qdisc_root_sleeping_lock(sch),
891 struct check_loop_arg
893 struct qdisc_walker w;
898 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
900 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
902 struct check_loop_arg arg;
904 if (q->ops->cl_ops == NULL)
907 arg.w.stop = arg.w.skip = arg.w.count = 0;
908 arg.w.fn = check_loop_fn;
911 q->ops->cl_ops->walk(q, &arg.w);
912 return arg.w.stop ? -ELOOP : 0;
916 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
919 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
920 struct check_loop_arg *arg = (struct check_loop_arg *)w;
922 leaf = cops->leaf(q, cl);
924 if (leaf == arg->p || arg->depth > 7)
926 return check_loop(leaf, arg->p, arg->depth + 1);
935 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
937 struct net *net = sock_net(skb->sk);
938 struct tcmsg *tcm = NLMSG_DATA(n);
939 struct nlattr *tca[TCA_MAX + 1];
940 struct net_device *dev;
941 u32 clid = tcm->tcm_parent;
942 struct Qdisc *q = NULL;
943 struct Qdisc *p = NULL;
946 if (net != &init_net)
949 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
952 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
957 if (clid != TC_H_ROOT) {
958 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
959 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
961 q = qdisc_leaf(p, clid);
962 } else { /* ingress */
963 q = dev->rx_queue.qdisc_sleeping;
966 struct netdev_queue *dev_queue;
967 dev_queue = netdev_get_tx_queue(dev, 0);
968 q = dev_queue->qdisc_sleeping;
973 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
976 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
980 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
983 if (n->nlmsg_type == RTM_DELQDISC) {
988 if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0)
991 qdisc_notify(skb, n, clid, NULL, q);
1000 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1002 struct net *net = sock_net(skb->sk);
1004 struct nlattr *tca[TCA_MAX + 1];
1005 struct net_device *dev;
1007 struct Qdisc *q, *p;
1010 if (net != &init_net)
1014 /* Reinit, just in case something touches this. */
1015 tcm = NLMSG_DATA(n);
1016 clid = tcm->tcm_parent;
1019 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1022 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1027 if (clid != TC_H_ROOT) {
1028 if (clid != TC_H_INGRESS) {
1029 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
1031 q = qdisc_leaf(p, clid);
1032 } else { /*ingress */
1033 q = dev->rx_queue.qdisc_sleeping;
1036 struct netdev_queue *dev_queue;
1037 dev_queue = netdev_get_tx_queue(dev, 0);
1038 q = dev_queue->qdisc_sleeping;
1041 /* It may be default qdisc, ignore it */
1042 if (q && q->handle == 0)
1045 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1046 if (tcm->tcm_handle) {
1047 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
1049 if (TC_H_MIN(tcm->tcm_handle))
1051 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
1052 goto create_n_graft;
1053 if (n->nlmsg_flags&NLM_F_EXCL)
1055 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1058 (p && check_loop(q, p, 0)))
1060 atomic_inc(&q->refcnt);
1064 goto create_n_graft;
1066 /* This magic test requires explanation.
1068 * We know, that some child q is already
1069 * attached to this parent and have choice:
1070 * either to change it or to create/graft new one.
1072 * 1. We are allowed to create/graft only
1073 * if CREATE and REPLACE flags are set.
1075 * 2. If EXCL is set, requestor wanted to say,
1076 * that qdisc tcm_handle is not expected
1077 * to exist, so that we choose create/graft too.
1079 * 3. The last case is when no flags are set.
1080 * Alas, it is sort of hole in API, we
1081 * cannot decide what to do unambiguously.
1082 * For now we select create/graft, if
1083 * user gave KIND, which does not match existing.
1085 if ((n->nlmsg_flags&NLM_F_CREATE) &&
1086 (n->nlmsg_flags&NLM_F_REPLACE) &&
1087 ((n->nlmsg_flags&NLM_F_EXCL) ||
1089 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1090 goto create_n_graft;
1094 if (!tcm->tcm_handle)
1096 q = qdisc_lookup(dev, tcm->tcm_handle);
1099 /* Change qdisc parameters */
1102 if (n->nlmsg_flags&NLM_F_EXCL)
1104 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1106 err = qdisc_change(q, tca);
1108 qdisc_notify(skb, n, clid, NULL, q);
1112 if (!(n->nlmsg_flags&NLM_F_CREATE))
1114 if (clid == TC_H_INGRESS)
1115 q = qdisc_create(dev, &dev->rx_queue,
1116 tcm->tcm_parent, tcm->tcm_parent,
1119 q = qdisc_create(dev, netdev_get_tx_queue(dev, 0),
1120 tcm->tcm_parent, tcm->tcm_handle,
1129 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1139 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1140 u32 pid, u32 seq, u16 flags, int event)
1143 struct nlmsghdr *nlh;
1144 unsigned char *b = skb_tail_pointer(skb);
1147 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1148 tcm = NLMSG_DATA(nlh);
1149 tcm->tcm_family = AF_UNSPEC;
1152 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1153 tcm->tcm_parent = clid;
1154 tcm->tcm_handle = q->handle;
1155 tcm->tcm_info = atomic_read(&q->refcnt);
1156 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1157 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1158 goto nla_put_failure;
1159 q->qstats.qlen = q->q.qlen;
1161 if (q->stab && qdisc_dump_stab(skb, q->stab) < 0)
1162 goto nla_put_failure;
1164 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1165 qdisc_root_sleeping_lock(q), &d) < 0)
1166 goto nla_put_failure;
1168 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1169 goto nla_put_failure;
1171 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1172 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
1173 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1174 goto nla_put_failure;
1176 if (gnet_stats_finish_copy(&d) < 0)
1177 goto nla_put_failure;
1179 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1188 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1189 u32 clid, struct Qdisc *old, struct Qdisc *new)
1191 struct sk_buff *skb;
1192 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1194 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1198 if (old && old->handle) {
1199 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
1203 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1208 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1215 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1217 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1220 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1221 struct netlink_callback *cb,
1222 int *q_idx_p, int s_q_idx)
1224 int ret = 0, q_idx = *q_idx_p;
1231 if (q_idx < s_q_idx) {
1234 if (!tc_qdisc_dump_ignore(q) &&
1235 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1236 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1240 list_for_each_entry(q, &root->list, list) {
1241 if (q_idx < s_q_idx) {
1245 if (!tc_qdisc_dump_ignore(q) &&
1246 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1247 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1260 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1262 struct net *net = sock_net(skb->sk);
1265 struct net_device *dev;
1267 if (net != &init_net)
1270 s_idx = cb->args[0];
1271 s_q_idx = q_idx = cb->args[1];
1272 read_lock(&dev_base_lock);
1274 for_each_netdev(&init_net, dev) {
1275 struct netdev_queue *dev_queue;
1283 dev_queue = netdev_get_tx_queue(dev, 0);
1284 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1287 dev_queue = &dev->rx_queue;
1288 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1296 read_unlock(&dev_base_lock);
1299 cb->args[1] = q_idx;
1306 /************************************************
1307 * Traffic classes manipulation. *
1308 ************************************************/
1312 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1314 struct net *net = sock_net(skb->sk);
1315 struct netdev_queue *dev_queue;
1316 struct tcmsg *tcm = NLMSG_DATA(n);
1317 struct nlattr *tca[TCA_MAX + 1];
1318 struct net_device *dev;
1319 struct Qdisc *q = NULL;
1320 const struct Qdisc_class_ops *cops;
1321 unsigned long cl = 0;
1322 unsigned long new_cl;
1323 u32 pid = tcm->tcm_parent;
1324 u32 clid = tcm->tcm_handle;
1325 u32 qid = TC_H_MAJ(clid);
1328 if (net != &init_net)
1331 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1334 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1339 parent == TC_H_UNSPEC - unspecified parent.
1340 parent == TC_H_ROOT - class is root, which has no parent.
1341 parent == X:0 - parent is root class.
1342 parent == X:Y - parent is a node in hierarchy.
1343 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1345 handle == 0:0 - generate handle from kernel pool.
1346 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1347 handle == X:Y - clear.
1348 handle == X:0 - root class.
1351 /* Step 1. Determine qdisc handle X:0 */
1353 dev_queue = netdev_get_tx_queue(dev, 0);
1354 if (pid != TC_H_ROOT) {
1355 u32 qid1 = TC_H_MAJ(pid);
1358 /* If both majors are known, they must be identical. */
1363 } else if (qid == 0)
1364 qid = dev_queue->qdisc_sleeping->handle;
1366 /* Now qid is genuine qdisc handle consistent
1367 both with parent and child.
1369 TC_H_MAJ(pid) still may be unspecified, complete it now.
1372 pid = TC_H_MAKE(qid, pid);
1375 qid = dev_queue->qdisc_sleeping->handle;
1378 /* OK. Locate qdisc */
1379 if ((q = qdisc_lookup(dev, qid)) == NULL)
1382 /* An check that it supports classes */
1383 cops = q->ops->cl_ops;
1387 /* Now try to get class */
1389 if (pid == TC_H_ROOT)
1392 clid = TC_H_MAKE(qid, clid);
1395 cl = cops->get(q, clid);
1399 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1402 switch (n->nlmsg_type) {
1405 if (n->nlmsg_flags&NLM_F_EXCL)
1409 err = cops->delete(q, cl);
1411 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1414 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1423 err = cops->change(q, clid, pid, tca, &new_cl);
1425 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1435 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1437 u32 pid, u32 seq, u16 flags, int event)
1440 struct nlmsghdr *nlh;
1441 unsigned char *b = skb_tail_pointer(skb);
1443 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1445 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1446 tcm = NLMSG_DATA(nlh);
1447 tcm->tcm_family = AF_UNSPEC;
1448 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1449 tcm->tcm_parent = q->handle;
1450 tcm->tcm_handle = q->handle;
1452 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1453 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1454 goto nla_put_failure;
1456 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1457 qdisc_root_sleeping_lock(q), &d) < 0)
1458 goto nla_put_failure;
1460 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1461 goto nla_put_failure;
1463 if (gnet_stats_finish_copy(&d) < 0)
1464 goto nla_put_failure;
1466 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1475 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1476 struct Qdisc *q, unsigned long cl, int event)
1478 struct sk_buff *skb;
1479 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1481 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1485 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1490 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1493 struct qdisc_dump_args
1495 struct qdisc_walker w;
1496 struct sk_buff *skb;
1497 struct netlink_callback *cb;
1500 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1502 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1504 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1505 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1508 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1509 struct tcmsg *tcm, struct netlink_callback *cb,
1512 struct qdisc_dump_args arg;
1514 if (tc_qdisc_dump_ignore(q) ||
1515 *t_p < s_t || !q->ops->cl_ops ||
1517 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1522 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1523 arg.w.fn = qdisc_class_dump;
1527 arg.w.skip = cb->args[1];
1529 q->ops->cl_ops->walk(q, &arg.w);
1530 cb->args[1] = arg.w.count;
1537 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1538 struct tcmsg *tcm, struct netlink_callback *cb,
1546 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1549 list_for_each_entry(q, &root->list, list) {
1550 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1557 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1559 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1560 struct net *net = sock_net(skb->sk);
1561 struct netdev_queue *dev_queue;
1562 struct net_device *dev;
1565 if (net != &init_net)
1568 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1570 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1576 dev_queue = netdev_get_tx_queue(dev, 0);
1577 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1580 dev_queue = &dev->rx_queue;
1581 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1591 /* Main classifier routine: scans classifier chain attached
1592 to this qdisc, (optionally) tests for protocol and asks
1593 specific classifiers.
1595 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1596 struct tcf_result *res)
1598 __be16 protocol = skb->protocol;
1601 for (; tp; tp = tp->next) {
1602 if ((tp->protocol == protocol ||
1603 tp->protocol == htons(ETH_P_ALL)) &&
1604 (err = tp->classify(skb, tp, res)) >= 0) {
1605 #ifdef CONFIG_NET_CLS_ACT
1606 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1607 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1614 EXPORT_SYMBOL(tc_classify_compat);
1616 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1617 struct tcf_result *res)
1621 #ifdef CONFIG_NET_CLS_ACT
1622 struct tcf_proto *otp = tp;
1625 protocol = skb->protocol;
1627 err = tc_classify_compat(skb, tp, res);
1628 #ifdef CONFIG_NET_CLS_ACT
1629 if (err == TC_ACT_RECLASSIFY) {
1630 u32 verd = G_TC_VERD(skb->tc_verd);
1633 if (verd++ >= MAX_REC_LOOP) {
1634 printk("rule prio %u protocol %02x reclassify loop, "
1636 tp->prio&0xffff, ntohs(tp->protocol));
1639 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1645 EXPORT_SYMBOL(tc_classify);
1647 void tcf_destroy(struct tcf_proto *tp)
1649 tp->ops->destroy(tp);
1650 module_put(tp->ops->owner);
1654 void tcf_destroy_chain(struct tcf_proto **fl)
1656 struct tcf_proto *tp;
1658 while ((tp = *fl) != NULL) {
1663 EXPORT_SYMBOL(tcf_destroy_chain);
1665 #ifdef CONFIG_PROC_FS
1666 static int psched_show(struct seq_file *seq, void *v)
1670 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1671 seq_printf(seq, "%08x %08x %08x %08x\n",
1672 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1674 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1679 static int psched_open(struct inode *inode, struct file *file)
1681 return single_open(file, psched_show, PDE(inode)->data);
1684 static const struct file_operations psched_fops = {
1685 .owner = THIS_MODULE,
1686 .open = psched_open,
1688 .llseek = seq_lseek,
1689 .release = single_release,
1693 static int __init pktsched_init(void)
1695 register_qdisc(&pfifo_qdisc_ops);
1696 register_qdisc(&bfifo_qdisc_ops);
1697 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1699 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1700 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1701 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1702 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1703 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1704 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1709 subsys_initcall(pktsched_init);