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 requeues once dequeued packet. It is used for non-standard or
103 just buggy devices, which can defer output even if netif_queue_stopped()=0.
107 returns qdisc to initial state: purge all buffers, clear all
108 timers, counters (except for statistics) etc.
112 initializes newly created qdisc.
116 destroys resources allocated by init and during lifetime of qdisc.
120 changes qdisc parameters.
123 /* Protects list of registered TC modules. It is pure SMP lock. */
124 static DEFINE_RWLOCK(qdisc_mod_lock);
127 /************************************************
128 * Queueing disciplines manipulation. *
129 ************************************************/
132 /* The list of all installed queueing disciplines. */
134 static struct Qdisc_ops *qdisc_base;
136 /* Register/uregister queueing discipline */
138 int register_qdisc(struct Qdisc_ops *qops)
140 struct Qdisc_ops *q, **qp;
143 write_lock(&qdisc_mod_lock);
144 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
145 if (!strcmp(qops->id, q->id))
148 if (qops->enqueue == NULL)
149 qops->enqueue = noop_qdisc_ops.enqueue;
150 if (qops->requeue == NULL)
151 qops->requeue = noop_qdisc_ops.requeue;
152 if (qops->dequeue == NULL)
153 qops->dequeue = noop_qdisc_ops.dequeue;
159 write_unlock(&qdisc_mod_lock);
162 EXPORT_SYMBOL(register_qdisc);
164 int unregister_qdisc(struct Qdisc_ops *qops)
166 struct Qdisc_ops *q, **qp;
169 write_lock(&qdisc_mod_lock);
170 for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next)
178 write_unlock(&qdisc_mod_lock);
181 EXPORT_SYMBOL(unregister_qdisc);
183 /* We know handle. Find qdisc among all qdisc's attached to device
184 (root qdisc, all its children, children of children etc.)
187 struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
191 if (!(root->flags & TCQ_F_BUILTIN) &&
192 root->handle == handle)
195 list_for_each_entry(q, &root->list, list) {
196 if (q->handle == handle)
202 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
206 for (i = 0; i < dev->num_tx_queues; i++) {
207 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
208 struct Qdisc *q, *txq_root = txq->qdisc_sleeping;
210 q = qdisc_match_from_root(txq_root, handle);
214 return qdisc_match_from_root(dev->rx_queue.qdisc_sleeping, handle);
217 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
221 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
225 cl = cops->get(p, classid);
229 leaf = cops->leaf(p, cl);
234 /* Find queueing discipline by name */
236 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
238 struct Qdisc_ops *q = NULL;
241 read_lock(&qdisc_mod_lock);
242 for (q = qdisc_base; q; q = q->next) {
243 if (nla_strcmp(kind, q->id) == 0) {
244 if (!try_module_get(q->owner))
249 read_unlock(&qdisc_mod_lock);
254 static struct qdisc_rate_table *qdisc_rtab_list;
256 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
258 struct qdisc_rate_table *rtab;
260 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
261 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
267 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
268 nla_len(tab) != TC_RTAB_SIZE)
271 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
275 memcpy(rtab->data, nla_data(tab), 1024);
276 rtab->next = qdisc_rtab_list;
277 qdisc_rtab_list = rtab;
281 EXPORT_SYMBOL(qdisc_get_rtab);
283 void qdisc_put_rtab(struct qdisc_rate_table *tab)
285 struct qdisc_rate_table *rtab, **rtabp;
287 if (!tab || --tab->refcnt)
290 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
298 EXPORT_SYMBOL(qdisc_put_rtab);
300 static LIST_HEAD(qdisc_stab_list);
301 static DEFINE_SPINLOCK(qdisc_stab_lock);
303 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
304 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
305 [TCA_STAB_DATA] = { .type = NLA_BINARY },
308 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
310 struct nlattr *tb[TCA_STAB_MAX + 1];
311 struct qdisc_size_table *stab;
312 struct tc_sizespec *s;
313 unsigned int tsize = 0;
317 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
320 if (!tb[TCA_STAB_BASE])
321 return ERR_PTR(-EINVAL);
323 s = nla_data(tb[TCA_STAB_BASE]);
326 if (!tb[TCA_STAB_DATA])
327 return ERR_PTR(-EINVAL);
328 tab = nla_data(tb[TCA_STAB_DATA]);
329 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
332 if (!s || tsize != s->tsize || (!tab && tsize > 0))
333 return ERR_PTR(-EINVAL);
335 spin_lock(&qdisc_stab_lock);
337 list_for_each_entry(stab, &qdisc_stab_list, list) {
338 if (memcmp(&stab->szopts, s, sizeof(*s)))
340 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
343 spin_unlock(&qdisc_stab_lock);
347 spin_unlock(&qdisc_stab_lock);
349 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
351 return ERR_PTR(-ENOMEM);
356 memcpy(stab->data, tab, tsize * sizeof(u16));
358 spin_lock(&qdisc_stab_lock);
359 list_add_tail(&stab->list, &qdisc_stab_list);
360 spin_unlock(&qdisc_stab_lock);
365 void qdisc_put_stab(struct qdisc_size_table *tab)
370 spin_lock(&qdisc_stab_lock);
372 if (--tab->refcnt == 0) {
373 list_del(&tab->list);
377 spin_unlock(&qdisc_stab_lock);
379 EXPORT_SYMBOL(qdisc_put_stab);
381 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
385 nest = nla_nest_start(skb, TCA_STAB);
386 NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts);
387 nla_nest_end(skb, nest);
395 void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab)
399 pkt_len = skb->len + stab->szopts.overhead;
400 if (unlikely(!stab->szopts.tsize))
403 slot = pkt_len + stab->szopts.cell_align;
404 if (unlikely(slot < 0))
407 slot >>= stab->szopts.cell_log;
408 if (likely(slot < stab->szopts.tsize))
409 pkt_len = stab->data[slot];
411 pkt_len = stab->data[stab->szopts.tsize - 1] *
412 (slot / stab->szopts.tsize) +
413 stab->data[slot % stab->szopts.tsize];
415 pkt_len <<= stab->szopts.size_log;
417 if (unlikely(pkt_len < 1))
419 qdisc_skb_cb(skb)->pkt_len = pkt_len;
421 EXPORT_SYMBOL(qdisc_calculate_pkt_len);
423 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
425 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
428 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
430 __netif_schedule(qdisc_root(wd->qdisc));
432 return HRTIMER_NORESTART;
435 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
437 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
438 wd->timer.function = qdisc_watchdog;
441 EXPORT_SYMBOL(qdisc_watchdog_init);
443 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
447 if (test_bit(__QDISC_STATE_DEACTIVATED,
448 &qdisc_root_sleeping(wd->qdisc)->state))
451 wd->qdisc->flags |= TCQ_F_THROTTLED;
452 time = ktime_set(0, 0);
453 time = ktime_add_ns(time, PSCHED_US2NS(expires));
454 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
456 EXPORT_SYMBOL(qdisc_watchdog_schedule);
458 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
460 hrtimer_cancel(&wd->timer);
461 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
463 EXPORT_SYMBOL(qdisc_watchdog_cancel);
465 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
467 unsigned int size = n * sizeof(struct hlist_head), i;
468 struct hlist_head *h;
470 if (size <= PAGE_SIZE)
471 h = kmalloc(size, GFP_KERNEL);
473 h = (struct hlist_head *)
474 __get_free_pages(GFP_KERNEL, get_order(size));
477 for (i = 0; i < n; i++)
478 INIT_HLIST_HEAD(&h[i]);
483 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
485 unsigned int size = n * sizeof(struct hlist_head);
487 if (size <= PAGE_SIZE)
490 free_pages((unsigned long)h, get_order(size));
493 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
495 struct Qdisc_class_common *cl;
496 struct hlist_node *n, *next;
497 struct hlist_head *nhash, *ohash;
498 unsigned int nsize, nmask, osize;
501 /* Rehash when load factor exceeds 0.75 */
502 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
504 nsize = clhash->hashsize * 2;
506 nhash = qdisc_class_hash_alloc(nsize);
510 ohash = clhash->hash;
511 osize = clhash->hashsize;
514 for (i = 0; i < osize; i++) {
515 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
516 h = qdisc_class_hash(cl->classid, nmask);
517 hlist_add_head(&cl->hnode, &nhash[h]);
520 clhash->hash = nhash;
521 clhash->hashsize = nsize;
522 clhash->hashmask = nmask;
523 sch_tree_unlock(sch);
525 qdisc_class_hash_free(ohash, osize);
527 EXPORT_SYMBOL(qdisc_class_hash_grow);
529 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
531 unsigned int size = 4;
533 clhash->hash = qdisc_class_hash_alloc(size);
534 if (clhash->hash == NULL)
536 clhash->hashsize = size;
537 clhash->hashmask = size - 1;
538 clhash->hashelems = 0;
541 EXPORT_SYMBOL(qdisc_class_hash_init);
543 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
545 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
547 EXPORT_SYMBOL(qdisc_class_hash_destroy);
549 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
550 struct Qdisc_class_common *cl)
554 INIT_HLIST_NODE(&cl->hnode);
555 h = qdisc_class_hash(cl->classid, clhash->hashmask);
556 hlist_add_head(&cl->hnode, &clhash->hash[h]);
559 EXPORT_SYMBOL(qdisc_class_hash_insert);
561 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
562 struct Qdisc_class_common *cl)
564 hlist_del(&cl->hnode);
567 EXPORT_SYMBOL(qdisc_class_hash_remove);
569 /* Allocate an unique handle from space managed by kernel */
571 static u32 qdisc_alloc_handle(struct net_device *dev)
574 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
577 autohandle += TC_H_MAKE(0x10000U, 0);
578 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
579 autohandle = TC_H_MAKE(0x80000000U, 0);
580 } while (qdisc_lookup(dev, autohandle) && --i > 0);
582 return i>0 ? autohandle : 0;
585 /* Attach toplevel qdisc to device queue. */
587 static struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
590 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
591 spinlock_t *root_lock;
593 root_lock = qdisc_root_lock(oqdisc);
594 spin_lock_bh(root_lock);
596 /* Prune old scheduler */
597 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
600 /* ... and graft new one */
603 dev_queue->qdisc_sleeping = qdisc;
604 dev_queue->qdisc = &noop_qdisc;
606 spin_unlock_bh(root_lock);
611 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
613 const struct Qdisc_class_ops *cops;
619 while ((parentid = sch->parent)) {
620 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
623 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
625 WARN_ON(parentid != TC_H_ROOT);
628 cops = sch->ops->cl_ops;
629 if (cops->qlen_notify) {
630 cl = cops->get(sch, parentid);
631 cops->qlen_notify(sch, cl);
637 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
639 static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid,
640 struct Qdisc *old, struct Qdisc *new)
643 qdisc_notify(skb, n, clid, old, new);
649 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
652 * When appropriate send a netlink notification using 'skb'
655 * On success, destroy old qdisc.
658 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
659 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
660 struct Qdisc *new, struct Qdisc *old)
662 struct Qdisc *q = old;
665 if (parent == NULL) {
666 unsigned int i, num_q, ingress;
669 num_q = dev->num_tx_queues;
670 if ((q && q->flags & TCQ_F_INGRESS) ||
671 (new && new->flags & TCQ_F_INGRESS)) {
676 if (dev->flags & IFF_UP)
679 for (i = 0; i < num_q; i++) {
680 struct netdev_queue *dev_queue = &dev->rx_queue;
683 dev_queue = netdev_get_tx_queue(dev, i);
685 old = dev_graft_qdisc(dev_queue, new);
687 atomic_inc(&new->refcnt);
689 notify_and_destroy(skb, n, classid, old, new);
692 if (dev->flags & IFF_UP)
695 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
700 unsigned long cl = cops->get(parent, classid);
702 err = cops->graft(parent, cl, new, &old);
703 cops->put(parent, cl);
707 notify_and_destroy(skb, n, classid, old, new);
712 /* lockdep annotation is needed for ingress; egress gets it only for name */
713 static struct lock_class_key qdisc_tx_lock;
714 static struct lock_class_key qdisc_rx_lock;
717 Allocate and initialize new qdisc.
719 Parameters are passed via opt.
722 static struct Qdisc *
723 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
724 u32 parent, u32 handle, struct nlattr **tca, int *errp)
727 struct nlattr *kind = tca[TCA_KIND];
729 struct Qdisc_ops *ops;
730 struct qdisc_size_table *stab;
732 ops = qdisc_lookup_ops(kind);
734 if (ops == NULL && kind != NULL) {
736 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
737 /* We dropped the RTNL semaphore in order to
738 * perform the module load. So, even if we
739 * succeeded in loading the module we have to
740 * tell the caller to replay the request. We
741 * indicate this using -EAGAIN.
742 * We replay the request because the device may
743 * go away in the mean time.
746 request_module("sch_%s", name);
748 ops = qdisc_lookup_ops(kind);
750 /* We will try again qdisc_lookup_ops,
751 * so don't keep a reference.
753 module_put(ops->owner);
765 sch = qdisc_alloc(dev_queue, ops);
771 sch->parent = parent;
773 if (handle == TC_H_INGRESS) {
774 sch->flags |= TCQ_F_INGRESS;
775 handle = TC_H_MAKE(TC_H_INGRESS, 0);
776 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
779 handle = qdisc_alloc_handle(dev);
784 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
787 sch->handle = handle;
789 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
791 stab = qdisc_get_stab(tca[TCA_STAB]);
799 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
800 qdisc_root_lock(sch),
804 * Any broken qdiscs that would require
805 * a ops->reset() here? The qdisc was never
806 * in action so it shouldn't be necessary.
813 if ((parent != TC_H_ROOT) && !(sch->flags & TCQ_F_INGRESS))
814 list_add_tail(&sch->list, &dev_queue->qdisc_sleeping->list);
819 qdisc_put_stab(sch->stab);
821 kfree((char *) sch - sch->padded);
823 module_put(ops->owner);
829 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
831 struct qdisc_size_table *stab = NULL;
834 if (tca[TCA_OPTIONS]) {
835 if (sch->ops->change == NULL)
837 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
843 stab = qdisc_get_stab(tca[TCA_STAB]);
845 return PTR_ERR(stab);
848 qdisc_put_stab(sch->stab);
852 gen_replace_estimator(&sch->bstats, &sch->rate_est,
853 qdisc_root_lock(sch), tca[TCA_RATE]);
857 struct check_loop_arg
859 struct qdisc_walker w;
864 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
866 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
868 struct check_loop_arg arg;
870 if (q->ops->cl_ops == NULL)
873 arg.w.stop = arg.w.skip = arg.w.count = 0;
874 arg.w.fn = check_loop_fn;
877 q->ops->cl_ops->walk(q, &arg.w);
878 return arg.w.stop ? -ELOOP : 0;
882 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
885 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
886 struct check_loop_arg *arg = (struct check_loop_arg *)w;
888 leaf = cops->leaf(q, cl);
890 if (leaf == arg->p || arg->depth > 7)
892 return check_loop(leaf, arg->p, arg->depth + 1);
901 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
903 struct net *net = sock_net(skb->sk);
904 struct tcmsg *tcm = NLMSG_DATA(n);
905 struct nlattr *tca[TCA_MAX + 1];
906 struct net_device *dev;
907 u32 clid = tcm->tcm_parent;
908 struct Qdisc *q = NULL;
909 struct Qdisc *p = NULL;
912 if (net != &init_net)
915 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
918 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
923 if (clid != TC_H_ROOT) {
924 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
925 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
927 q = qdisc_leaf(p, clid);
928 } else { /* ingress */
929 q = dev->rx_queue.qdisc_sleeping;
932 struct netdev_queue *dev_queue;
933 dev_queue = netdev_get_tx_queue(dev, 0);
934 q = dev_queue->qdisc_sleeping;
939 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
942 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
946 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
949 if (n->nlmsg_type == RTM_DELQDISC) {
954 if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0)
957 qdisc_notify(skb, n, clid, NULL, q);
966 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
968 struct net *net = sock_net(skb->sk);
970 struct nlattr *tca[TCA_MAX + 1];
971 struct net_device *dev;
976 if (net != &init_net)
980 /* Reinit, just in case something touches this. */
982 clid = tcm->tcm_parent;
985 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
988 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
993 if (clid != TC_H_ROOT) {
994 if (clid != TC_H_INGRESS) {
995 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
997 q = qdisc_leaf(p, clid);
998 } else { /*ingress */
999 q = dev->rx_queue.qdisc_sleeping;
1002 struct netdev_queue *dev_queue;
1003 dev_queue = netdev_get_tx_queue(dev, 0);
1004 q = dev_queue->qdisc_sleeping;
1007 /* It may be default qdisc, ignore it */
1008 if (q && q->handle == 0)
1011 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1012 if (tcm->tcm_handle) {
1013 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
1015 if (TC_H_MIN(tcm->tcm_handle))
1017 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
1018 goto create_n_graft;
1019 if (n->nlmsg_flags&NLM_F_EXCL)
1021 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1024 (p && check_loop(q, p, 0)))
1026 atomic_inc(&q->refcnt);
1030 goto create_n_graft;
1032 /* This magic test requires explanation.
1034 * We know, that some child q is already
1035 * attached to this parent and have choice:
1036 * either to change it or to create/graft new one.
1038 * 1. We are allowed to create/graft only
1039 * if CREATE and REPLACE flags are set.
1041 * 2. If EXCL is set, requestor wanted to say,
1042 * that qdisc tcm_handle is not expected
1043 * to exist, so that we choose create/graft too.
1045 * 3. The last case is when no flags are set.
1046 * Alas, it is sort of hole in API, we
1047 * cannot decide what to do unambiguously.
1048 * For now we select create/graft, if
1049 * user gave KIND, which does not match existing.
1051 if ((n->nlmsg_flags&NLM_F_CREATE) &&
1052 (n->nlmsg_flags&NLM_F_REPLACE) &&
1053 ((n->nlmsg_flags&NLM_F_EXCL) ||
1055 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1056 goto create_n_graft;
1060 if (!tcm->tcm_handle)
1062 q = qdisc_lookup(dev, tcm->tcm_handle);
1065 /* Change qdisc parameters */
1068 if (n->nlmsg_flags&NLM_F_EXCL)
1070 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1072 err = qdisc_change(q, tca);
1074 qdisc_notify(skb, n, clid, NULL, q);
1078 if (!(n->nlmsg_flags&NLM_F_CREATE))
1080 if (clid == TC_H_INGRESS)
1081 q = qdisc_create(dev, &dev->rx_queue,
1082 tcm->tcm_parent, tcm->tcm_parent,
1085 q = qdisc_create(dev, netdev_get_tx_queue(dev, 0),
1086 tcm->tcm_parent, tcm->tcm_handle,
1095 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1105 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1106 u32 pid, u32 seq, u16 flags, int event)
1109 struct nlmsghdr *nlh;
1110 unsigned char *b = skb_tail_pointer(skb);
1113 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1114 tcm = NLMSG_DATA(nlh);
1115 tcm->tcm_family = AF_UNSPEC;
1118 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1119 tcm->tcm_parent = clid;
1120 tcm->tcm_handle = q->handle;
1121 tcm->tcm_info = atomic_read(&q->refcnt);
1122 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1123 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1124 goto nla_put_failure;
1125 q->qstats.qlen = q->q.qlen;
1127 if (q->stab && qdisc_dump_stab(skb, q->stab) < 0)
1128 goto nla_put_failure;
1130 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1131 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
1132 goto nla_put_failure;
1134 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1135 goto nla_put_failure;
1137 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1138 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
1139 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1140 goto nla_put_failure;
1142 if (gnet_stats_finish_copy(&d) < 0)
1143 goto nla_put_failure;
1145 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1154 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1155 u32 clid, struct Qdisc *old, struct Qdisc *new)
1157 struct sk_buff *skb;
1158 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1160 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1164 if (old && old->handle) {
1165 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
1169 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1174 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1181 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1183 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1186 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1187 struct netlink_callback *cb,
1188 int *q_idx_p, int s_q_idx)
1190 int ret = 0, q_idx = *q_idx_p;
1197 if (q_idx < s_q_idx) {
1200 if (!tc_qdisc_dump_ignore(q) &&
1201 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1202 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1206 list_for_each_entry(q, &root->list, list) {
1207 if (q_idx < s_q_idx) {
1211 if (!tc_qdisc_dump_ignore(q) &&
1212 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1213 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1226 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1228 struct net *net = sock_net(skb->sk);
1231 struct net_device *dev;
1233 if (net != &init_net)
1236 s_idx = cb->args[0];
1237 s_q_idx = q_idx = cb->args[1];
1238 read_lock(&dev_base_lock);
1240 for_each_netdev(&init_net, dev) {
1241 struct netdev_queue *dev_queue;
1249 dev_queue = netdev_get_tx_queue(dev, 0);
1250 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1253 dev_queue = &dev->rx_queue;
1254 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1262 read_unlock(&dev_base_lock);
1265 cb->args[1] = q_idx;
1272 /************************************************
1273 * Traffic classes manipulation. *
1274 ************************************************/
1278 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1280 struct net *net = sock_net(skb->sk);
1281 struct netdev_queue *dev_queue;
1282 struct tcmsg *tcm = NLMSG_DATA(n);
1283 struct nlattr *tca[TCA_MAX + 1];
1284 struct net_device *dev;
1285 struct Qdisc *q = NULL;
1286 const struct Qdisc_class_ops *cops;
1287 unsigned long cl = 0;
1288 unsigned long new_cl;
1289 u32 pid = tcm->tcm_parent;
1290 u32 clid = tcm->tcm_handle;
1291 u32 qid = TC_H_MAJ(clid);
1294 if (net != &init_net)
1297 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1300 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1305 parent == TC_H_UNSPEC - unspecified parent.
1306 parent == TC_H_ROOT - class is root, which has no parent.
1307 parent == X:0 - parent is root class.
1308 parent == X:Y - parent is a node in hierarchy.
1309 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1311 handle == 0:0 - generate handle from kernel pool.
1312 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1313 handle == X:Y - clear.
1314 handle == X:0 - root class.
1317 /* Step 1. Determine qdisc handle X:0 */
1319 dev_queue = netdev_get_tx_queue(dev, 0);
1320 if (pid != TC_H_ROOT) {
1321 u32 qid1 = TC_H_MAJ(pid);
1324 /* If both majors are known, they must be identical. */
1329 } else if (qid == 0)
1330 qid = dev_queue->qdisc_sleeping->handle;
1332 /* Now qid is genuine qdisc handle consistent
1333 both with parent and child.
1335 TC_H_MAJ(pid) still may be unspecified, complete it now.
1338 pid = TC_H_MAKE(qid, pid);
1341 qid = dev_queue->qdisc_sleeping->handle;
1344 /* OK. Locate qdisc */
1345 if ((q = qdisc_lookup(dev, qid)) == NULL)
1348 /* An check that it supports classes */
1349 cops = q->ops->cl_ops;
1353 /* Now try to get class */
1355 if (pid == TC_H_ROOT)
1358 clid = TC_H_MAKE(qid, clid);
1361 cl = cops->get(q, clid);
1365 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1368 switch (n->nlmsg_type) {
1371 if (n->nlmsg_flags&NLM_F_EXCL)
1375 err = cops->delete(q, cl);
1377 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1380 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1389 err = cops->change(q, clid, pid, tca, &new_cl);
1391 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1401 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1403 u32 pid, u32 seq, u16 flags, int event)
1406 struct nlmsghdr *nlh;
1407 unsigned char *b = skb_tail_pointer(skb);
1409 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1411 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1412 tcm = NLMSG_DATA(nlh);
1413 tcm->tcm_family = AF_UNSPEC;
1414 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1415 tcm->tcm_parent = q->handle;
1416 tcm->tcm_handle = q->handle;
1418 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1419 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1420 goto nla_put_failure;
1422 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1423 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
1424 goto nla_put_failure;
1426 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1427 goto nla_put_failure;
1429 if (gnet_stats_finish_copy(&d) < 0)
1430 goto nla_put_failure;
1432 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1441 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1442 struct Qdisc *q, unsigned long cl, int event)
1444 struct sk_buff *skb;
1445 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1447 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1451 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1456 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1459 struct qdisc_dump_args
1461 struct qdisc_walker w;
1462 struct sk_buff *skb;
1463 struct netlink_callback *cb;
1466 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1468 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1470 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1471 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1474 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1475 struct tcmsg *tcm, struct netlink_callback *cb,
1478 struct qdisc_dump_args arg;
1480 if (tc_qdisc_dump_ignore(q) ||
1481 *t_p < s_t || !q->ops->cl_ops ||
1483 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1488 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1489 arg.w.fn = qdisc_class_dump;
1493 arg.w.skip = cb->args[1];
1495 q->ops->cl_ops->walk(q, &arg.w);
1496 cb->args[1] = arg.w.count;
1503 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1504 struct tcmsg *tcm, struct netlink_callback *cb,
1512 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1515 list_for_each_entry(q, &root->list, list) {
1516 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1523 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1525 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1526 struct net *net = sock_net(skb->sk);
1527 struct netdev_queue *dev_queue;
1528 struct net_device *dev;
1531 if (net != &init_net)
1534 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1536 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1542 dev_queue = netdev_get_tx_queue(dev, 0);
1543 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1546 dev_queue = &dev->rx_queue;
1547 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1557 /* Main classifier routine: scans classifier chain attached
1558 to this qdisc, (optionally) tests for protocol and asks
1559 specific classifiers.
1561 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1562 struct tcf_result *res)
1564 __be16 protocol = skb->protocol;
1567 for (; tp; tp = tp->next) {
1568 if ((tp->protocol == protocol ||
1569 tp->protocol == htons(ETH_P_ALL)) &&
1570 (err = tp->classify(skb, tp, res)) >= 0) {
1571 #ifdef CONFIG_NET_CLS_ACT
1572 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1573 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1580 EXPORT_SYMBOL(tc_classify_compat);
1582 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1583 struct tcf_result *res)
1587 #ifdef CONFIG_NET_CLS_ACT
1588 struct tcf_proto *otp = tp;
1591 protocol = skb->protocol;
1593 err = tc_classify_compat(skb, tp, res);
1594 #ifdef CONFIG_NET_CLS_ACT
1595 if (err == TC_ACT_RECLASSIFY) {
1596 u32 verd = G_TC_VERD(skb->tc_verd);
1599 if (verd++ >= MAX_REC_LOOP) {
1600 printk("rule prio %u protocol %02x reclassify loop, "
1602 tp->prio&0xffff, ntohs(tp->protocol));
1605 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1611 EXPORT_SYMBOL(tc_classify);
1613 void tcf_destroy(struct tcf_proto *tp)
1615 tp->ops->destroy(tp);
1616 module_put(tp->ops->owner);
1620 void tcf_destroy_chain(struct tcf_proto **fl)
1622 struct tcf_proto *tp;
1624 while ((tp = *fl) != NULL) {
1629 EXPORT_SYMBOL(tcf_destroy_chain);
1631 #ifdef CONFIG_PROC_FS
1632 static int psched_show(struct seq_file *seq, void *v)
1636 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1637 seq_printf(seq, "%08x %08x %08x %08x\n",
1638 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1640 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1645 static int psched_open(struct inode *inode, struct file *file)
1647 return single_open(file, psched_show, PDE(inode)->data);
1650 static const struct file_operations psched_fops = {
1651 .owner = THIS_MODULE,
1652 .open = psched_open,
1654 .llseek = seq_lseek,
1655 .release = single_release,
1659 static int __init pktsched_init(void)
1661 register_qdisc(&pfifo_qdisc_ops);
1662 register_qdisc(&bfifo_qdisc_ops);
1663 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1665 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1666 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1667 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1668 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1669 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1670 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1675 subsys_initcall(pktsched_init);