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>
31 #include <net/net_namespace.h>
33 #include <net/netlink.h>
34 #include <net/pkt_sched.h>
36 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid,
37 struct Qdisc *old, struct Qdisc *new);
38 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
39 struct Qdisc *q, unsigned long cl, int event);
46 This file consists of two interrelated parts:
48 1. queueing disciplines manager frontend.
49 2. traffic classes manager frontend.
51 Generally, queueing discipline ("qdisc") is a black box,
52 which is able to enqueue packets and to dequeue them (when
53 device is ready to send something) in order and at times
54 determined by algorithm hidden in it.
56 qdisc's are divided to two categories:
57 - "queues", which have no internal structure visible from outside.
58 - "schedulers", which split all the packets to "traffic classes",
59 using "packet classifiers" (look at cls_api.c)
61 In turn, classes may have child qdiscs (as rule, queues)
62 attached to them etc. etc. etc.
64 The goal of the routines in this file is to translate
65 information supplied by user in the form of handles
66 to more intelligible for kernel form, to make some sanity
67 checks and part of work, which is common to all qdiscs
68 and to provide rtnetlink notifications.
70 All real intelligent work is done inside qdisc modules.
74 Every discipline has two major routines: enqueue and dequeue.
78 dequeue usually returns a skb to send. It is allowed to return NULL,
79 but it does not mean that queue is empty, it just means that
80 discipline does not want to send anything this time.
81 Queue is really empty if q->q.qlen == 0.
82 For complicated disciplines with multiple queues q->q is not
83 real packet queue, but however q->q.qlen must be valid.
87 enqueue returns 0, if packet was enqueued successfully.
88 If packet (this one or another one) was dropped, it returns
90 NET_XMIT_DROP - this packet dropped
91 Expected action: do not backoff, but wait until queue will clear.
92 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
93 Expected action: backoff or ignore
94 NET_XMIT_POLICED - dropped by police.
95 Expected action: backoff or error to real-time apps.
101 requeues once dequeued packet. It is used for non-standard or
102 just buggy devices, which can defer output even if netif_queue_stopped()=0.
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->requeue == NULL)
150 qops->requeue = noop_qdisc_ops.requeue;
151 if (qops->dequeue == NULL)
152 qops->dequeue = noop_qdisc_ops.dequeue;
158 write_unlock(&qdisc_mod_lock);
161 EXPORT_SYMBOL(register_qdisc);
163 int unregister_qdisc(struct Qdisc_ops *qops)
165 struct Qdisc_ops *q, **qp;
168 write_lock(&qdisc_mod_lock);
169 for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next)
177 write_unlock(&qdisc_mod_lock);
180 EXPORT_SYMBOL(unregister_qdisc);
182 /* We know handle. Find qdisc among all qdisc's attached to device
183 (root qdisc, all its children, children of children etc.)
186 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
188 struct netdev_queue *dev_queue = &dev->tx_queue;
191 list_for_each_entry(q, &dev_queue->qdisc_list, list) {
192 if (q->handle == handle)
198 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
202 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
206 cl = cops->get(p, classid);
210 leaf = cops->leaf(p, cl);
215 /* Find queueing discipline by name */
217 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
219 struct Qdisc_ops *q = NULL;
222 read_lock(&qdisc_mod_lock);
223 for (q = qdisc_base; q; q = q->next) {
224 if (nla_strcmp(kind, q->id) == 0) {
225 if (!try_module_get(q->owner))
230 read_unlock(&qdisc_mod_lock);
235 static struct qdisc_rate_table *qdisc_rtab_list;
237 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
239 struct qdisc_rate_table *rtab;
241 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
242 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
248 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
249 nla_len(tab) != TC_RTAB_SIZE)
252 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
256 memcpy(rtab->data, nla_data(tab), 1024);
257 rtab->next = qdisc_rtab_list;
258 qdisc_rtab_list = rtab;
262 EXPORT_SYMBOL(qdisc_get_rtab);
264 void qdisc_put_rtab(struct qdisc_rate_table *tab)
266 struct qdisc_rate_table *rtab, **rtabp;
268 if (!tab || --tab->refcnt)
271 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
279 EXPORT_SYMBOL(qdisc_put_rtab);
281 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
283 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
285 struct netdev_queue *txq = wd->qdisc->dev_queue;
287 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
289 netif_schedule_queue(txq);
291 return HRTIMER_NORESTART;
294 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
296 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
297 wd->timer.function = qdisc_watchdog;
300 EXPORT_SYMBOL(qdisc_watchdog_init);
302 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
306 wd->qdisc->flags |= TCQ_F_THROTTLED;
307 time = ktime_set(0, 0);
308 time = ktime_add_ns(time, PSCHED_US2NS(expires));
309 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
311 EXPORT_SYMBOL(qdisc_watchdog_schedule);
313 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
315 hrtimer_cancel(&wd->timer);
316 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
318 EXPORT_SYMBOL(qdisc_watchdog_cancel);
320 struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
322 unsigned int size = n * sizeof(struct hlist_head), i;
323 struct hlist_head *h;
325 if (size <= PAGE_SIZE)
326 h = kmalloc(size, GFP_KERNEL);
328 h = (struct hlist_head *)
329 __get_free_pages(GFP_KERNEL, get_order(size));
332 for (i = 0; i < n; i++)
333 INIT_HLIST_HEAD(&h[i]);
338 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
340 unsigned int size = n * sizeof(struct hlist_head);
342 if (size <= PAGE_SIZE)
345 free_pages((unsigned long)h, get_order(size));
348 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
350 struct Qdisc_class_common *cl;
351 struct hlist_node *n, *next;
352 struct hlist_head *nhash, *ohash;
353 unsigned int nsize, nmask, osize;
356 /* Rehash when load factor exceeds 0.75 */
357 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
359 nsize = clhash->hashsize * 2;
361 nhash = qdisc_class_hash_alloc(nsize);
365 ohash = clhash->hash;
366 osize = clhash->hashsize;
369 for (i = 0; i < osize; i++) {
370 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
371 h = qdisc_class_hash(cl->classid, nmask);
372 hlist_add_head(&cl->hnode, &nhash[h]);
375 clhash->hash = nhash;
376 clhash->hashsize = nsize;
377 clhash->hashmask = nmask;
378 sch_tree_unlock(sch);
380 qdisc_class_hash_free(ohash, osize);
382 EXPORT_SYMBOL(qdisc_class_hash_grow);
384 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
386 unsigned int size = 4;
388 clhash->hash = qdisc_class_hash_alloc(size);
389 if (clhash->hash == NULL)
391 clhash->hashsize = size;
392 clhash->hashmask = size - 1;
393 clhash->hashelems = 0;
396 EXPORT_SYMBOL(qdisc_class_hash_init);
398 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
400 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
402 EXPORT_SYMBOL(qdisc_class_hash_destroy);
404 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
405 struct Qdisc_class_common *cl)
409 INIT_HLIST_NODE(&cl->hnode);
410 h = qdisc_class_hash(cl->classid, clhash->hashmask);
411 hlist_add_head(&cl->hnode, &clhash->hash[h]);
414 EXPORT_SYMBOL(qdisc_class_hash_insert);
416 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
417 struct Qdisc_class_common *cl)
419 hlist_del(&cl->hnode);
422 EXPORT_SYMBOL(qdisc_class_hash_remove);
424 /* Allocate an unique handle from space managed by kernel */
426 static u32 qdisc_alloc_handle(struct net_device *dev)
429 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
432 autohandle += TC_H_MAKE(0x10000U, 0);
433 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
434 autohandle = TC_H_MAKE(0x80000000U, 0);
435 } while (qdisc_lookup(dev, autohandle) && --i > 0);
437 return i>0 ? autohandle : 0;
440 /* Attach toplevel qdisc to device dev */
442 static struct Qdisc *
443 dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc)
445 struct netdev_queue *dev_queue;
446 struct Qdisc *oqdisc;
448 if (dev->flags & IFF_UP)
451 qdisc_lock_tree(dev);
452 if (qdisc && qdisc->flags&TCQ_F_INGRESS) {
453 dev_queue = &dev->rx_queue;
454 oqdisc = dev_queue->qdisc;
455 /* Prune old scheduler */
456 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) {
459 dev_queue->qdisc = NULL;
461 dev_queue->qdisc = qdisc;
465 dev_queue = &dev->tx_queue;
466 oqdisc = dev_queue->qdisc_sleeping;
468 /* Prune old scheduler */
469 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
472 /* ... and graft new one */
475 dev_queue->qdisc_sleeping = qdisc;
476 dev_queue->qdisc = &noop_qdisc;
479 qdisc_unlock_tree(dev);
481 if (dev->flags & IFF_UP)
487 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
489 const struct Qdisc_class_ops *cops;
495 while ((parentid = sch->parent)) {
496 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
499 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
501 WARN_ON(parentid != TC_H_ROOT);
504 cops = sch->ops->cl_ops;
505 if (cops->qlen_notify) {
506 cl = cops->get(sch, parentid);
507 cops->qlen_notify(sch, cl);
513 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
515 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
518 Old qdisc is not destroyed but returned in *old.
521 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
523 struct Qdisc *new, struct Qdisc **old)
526 struct Qdisc *q = *old;
529 if (parent == NULL) {
530 if (q && q->flags&TCQ_F_INGRESS) {
531 *old = dev_graft_qdisc(dev, q);
533 *old = dev_graft_qdisc(dev, new);
536 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
541 unsigned long cl = cops->get(parent, classid);
543 err = cops->graft(parent, cl, new, old);
544 cops->put(parent, cl);
552 Allocate and initialize new qdisc.
554 Parameters are passed via opt.
557 static struct Qdisc *
558 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
559 u32 parent, u32 handle, struct nlattr **tca, int *errp)
562 struct nlattr *kind = tca[TCA_KIND];
564 struct Qdisc_ops *ops;
566 ops = qdisc_lookup_ops(kind);
568 if (ops == NULL && kind != NULL) {
570 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
571 /* We dropped the RTNL semaphore in order to
572 * perform the module load. So, even if we
573 * succeeded in loading the module we have to
574 * tell the caller to replay the request. We
575 * indicate this using -EAGAIN.
576 * We replay the request because the device may
577 * go away in the mean time.
580 request_module("sch_%s", name);
582 ops = qdisc_lookup_ops(kind);
584 /* We will try again qdisc_lookup_ops,
585 * so don't keep a reference.
587 module_put(ops->owner);
599 sch = qdisc_alloc(dev_queue, ops);
605 sch->parent = parent;
607 if (handle == TC_H_INGRESS) {
608 sch->flags |= TCQ_F_INGRESS;
609 handle = TC_H_MAKE(TC_H_INGRESS, 0);
612 handle = qdisc_alloc_handle(dev);
619 sch->handle = handle;
621 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
623 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
624 &sch->dev_queue->lock,
628 * Any broken qdiscs that would require
629 * a ops->reset() here? The qdisc was never
630 * in action so it shouldn't be necessary.
637 qdisc_lock_tree(dev);
638 list_add_tail(&sch->list, &dev_queue->qdisc_list);
639 qdisc_unlock_tree(dev);
645 kfree((char *) sch - sch->padded);
647 module_put(ops->owner);
653 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
655 if (tca[TCA_OPTIONS]) {
658 if (sch->ops->change == NULL)
660 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
665 gen_replace_estimator(&sch->bstats, &sch->rate_est,
666 &sch->dev_queue->lock, tca[TCA_RATE]);
670 struct check_loop_arg
672 struct qdisc_walker w;
677 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
679 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
681 struct check_loop_arg arg;
683 if (q->ops->cl_ops == NULL)
686 arg.w.stop = arg.w.skip = arg.w.count = 0;
687 arg.w.fn = check_loop_fn;
690 q->ops->cl_ops->walk(q, &arg.w);
691 return arg.w.stop ? -ELOOP : 0;
695 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
698 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
699 struct check_loop_arg *arg = (struct check_loop_arg *)w;
701 leaf = cops->leaf(q, cl);
703 if (leaf == arg->p || arg->depth > 7)
705 return check_loop(leaf, arg->p, arg->depth + 1);
714 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
716 struct net *net = sock_net(skb->sk);
717 struct tcmsg *tcm = NLMSG_DATA(n);
718 struct nlattr *tca[TCA_MAX + 1];
719 struct net_device *dev;
720 u32 clid = tcm->tcm_parent;
721 struct Qdisc *q = NULL;
722 struct Qdisc *p = NULL;
725 if (net != &init_net)
728 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
731 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
736 if (clid != TC_H_ROOT) {
737 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
738 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
740 q = qdisc_leaf(p, clid);
741 } else { /* ingress */
742 q = dev->rx_queue.qdisc;
745 struct netdev_queue *dev_queue = &dev->tx_queue;
746 q = dev_queue->qdisc_sleeping;
751 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
754 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
758 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
761 if (n->nlmsg_type == RTM_DELQDISC) {
766 if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0)
769 qdisc_notify(skb, n, clid, q, NULL);
770 qdisc_lock_tree(dev);
772 qdisc_unlock_tree(dev);
775 qdisc_notify(skb, n, clid, NULL, q);
784 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
786 struct net *net = sock_net(skb->sk);
788 struct nlattr *tca[TCA_MAX + 1];
789 struct net_device *dev;
794 if (net != &init_net)
798 /* Reinit, just in case something touches this. */
800 clid = tcm->tcm_parent;
803 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
806 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
811 if (clid != TC_H_ROOT) {
812 if (clid != TC_H_INGRESS) {
813 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
815 q = qdisc_leaf(p, clid);
816 } else { /*ingress */
817 q = dev->rx_queue.qdisc;
820 struct netdev_queue *dev_queue = &dev->tx_queue;
821 q = dev_queue->qdisc_sleeping;
824 /* It may be default qdisc, ignore it */
825 if (q && q->handle == 0)
828 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
829 if (tcm->tcm_handle) {
830 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
832 if (TC_H_MIN(tcm->tcm_handle))
834 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
836 if (n->nlmsg_flags&NLM_F_EXCL)
838 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
841 (p && check_loop(q, p, 0)))
843 atomic_inc(&q->refcnt);
849 /* This magic test requires explanation.
851 * We know, that some child q is already
852 * attached to this parent and have choice:
853 * either to change it or to create/graft new one.
855 * 1. We are allowed to create/graft only
856 * if CREATE and REPLACE flags are set.
858 * 2. If EXCL is set, requestor wanted to say,
859 * that qdisc tcm_handle is not expected
860 * to exist, so that we choose create/graft too.
862 * 3. The last case is when no flags are set.
863 * Alas, it is sort of hole in API, we
864 * cannot decide what to do unambiguously.
865 * For now we select create/graft, if
866 * user gave KIND, which does not match existing.
868 if ((n->nlmsg_flags&NLM_F_CREATE) &&
869 (n->nlmsg_flags&NLM_F_REPLACE) &&
870 ((n->nlmsg_flags&NLM_F_EXCL) ||
872 nla_strcmp(tca[TCA_KIND], q->ops->id))))
877 if (!tcm->tcm_handle)
879 q = qdisc_lookup(dev, tcm->tcm_handle);
882 /* Change qdisc parameters */
885 if (n->nlmsg_flags&NLM_F_EXCL)
887 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
889 err = qdisc_change(q, tca);
891 qdisc_notify(skb, n, clid, NULL, q);
895 if (!(n->nlmsg_flags&NLM_F_CREATE))
897 if (clid == TC_H_INGRESS)
898 q = qdisc_create(dev, &dev->rx_queue,
899 tcm->tcm_parent, tcm->tcm_parent,
902 q = qdisc_create(dev, &dev->tx_queue,
903 tcm->tcm_parent, tcm->tcm_handle,
913 struct Qdisc *old_q = NULL;
914 err = qdisc_graft(dev, p, clid, q, &old_q);
917 qdisc_lock_tree(dev);
919 qdisc_unlock_tree(dev);
923 qdisc_notify(skb, n, clid, old_q, q);
925 qdisc_lock_tree(dev);
926 qdisc_destroy(old_q);
927 qdisc_unlock_tree(dev);
933 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
934 u32 pid, u32 seq, u16 flags, int event)
937 struct nlmsghdr *nlh;
938 unsigned char *b = skb_tail_pointer(skb);
941 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
942 tcm = NLMSG_DATA(nlh);
943 tcm->tcm_family = AF_UNSPEC;
946 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
947 tcm->tcm_parent = clid;
948 tcm->tcm_handle = q->handle;
949 tcm->tcm_info = atomic_read(&q->refcnt);
950 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
951 if (q->ops->dump && q->ops->dump(q, skb) < 0)
952 goto nla_put_failure;
953 q->qstats.qlen = q->q.qlen;
955 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
956 TCA_XSTATS, &q->dev_queue->lock, &d) < 0)
957 goto nla_put_failure;
959 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
960 goto nla_put_failure;
962 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
963 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
964 gnet_stats_copy_queue(&d, &q->qstats) < 0)
965 goto nla_put_failure;
967 if (gnet_stats_finish_copy(&d) < 0)
968 goto nla_put_failure;
970 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
979 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
980 u32 clid, struct Qdisc *old, struct Qdisc *new)
983 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
985 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
989 if (old && old->handle) {
990 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
994 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
999 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1006 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1008 struct net *net = sock_net(skb->sk);
1011 struct net_device *dev;
1014 if (net != &init_net)
1017 s_idx = cb->args[0];
1018 s_q_idx = q_idx = cb->args[1];
1019 read_lock(&dev_base_lock);
1021 for_each_netdev(&init_net, dev) {
1022 struct netdev_queue *dev_queue;
1028 dev_queue = &dev->tx_queue;
1029 list_for_each_entry(q, &dev_queue->qdisc_list, list) {
1030 if (q_idx < s_q_idx) {
1034 if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1035 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1044 read_unlock(&dev_base_lock);
1047 cb->args[1] = q_idx;
1054 /************************************************
1055 * Traffic classes manipulation. *
1056 ************************************************/
1060 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1062 struct net *net = sock_net(skb->sk);
1063 struct netdev_queue *dev_queue;
1064 struct tcmsg *tcm = NLMSG_DATA(n);
1065 struct nlattr *tca[TCA_MAX + 1];
1066 struct net_device *dev;
1067 struct Qdisc *q = NULL;
1068 const struct Qdisc_class_ops *cops;
1069 unsigned long cl = 0;
1070 unsigned long new_cl;
1071 u32 pid = tcm->tcm_parent;
1072 u32 clid = tcm->tcm_handle;
1073 u32 qid = TC_H_MAJ(clid);
1076 if (net != &init_net)
1079 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1082 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1087 parent == TC_H_UNSPEC - unspecified parent.
1088 parent == TC_H_ROOT - class is root, which has no parent.
1089 parent == X:0 - parent is root class.
1090 parent == X:Y - parent is a node in hierarchy.
1091 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1093 handle == 0:0 - generate handle from kernel pool.
1094 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1095 handle == X:Y - clear.
1096 handle == X:0 - root class.
1099 /* Step 1. Determine qdisc handle X:0 */
1101 dev_queue = &dev->tx_queue;
1102 if (pid != TC_H_ROOT) {
1103 u32 qid1 = TC_H_MAJ(pid);
1106 /* If both majors are known, they must be identical. */
1111 } else if (qid == 0)
1112 qid = dev_queue->qdisc_sleeping->handle;
1114 /* Now qid is genuine qdisc handle consistent
1115 both with parent and child.
1117 TC_H_MAJ(pid) still may be unspecified, complete it now.
1120 pid = TC_H_MAKE(qid, pid);
1123 qid = dev_queue->qdisc_sleeping->handle;
1126 /* OK. Locate qdisc */
1127 if ((q = qdisc_lookup(dev, qid)) == NULL)
1130 /* An check that it supports classes */
1131 cops = q->ops->cl_ops;
1135 /* Now try to get class */
1137 if (pid == TC_H_ROOT)
1140 clid = TC_H_MAKE(qid, clid);
1143 cl = cops->get(q, clid);
1147 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1150 switch (n->nlmsg_type) {
1153 if (n->nlmsg_flags&NLM_F_EXCL)
1157 err = cops->delete(q, cl);
1159 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1162 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1171 err = cops->change(q, clid, pid, tca, &new_cl);
1173 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1183 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1185 u32 pid, u32 seq, u16 flags, int event)
1188 struct nlmsghdr *nlh;
1189 unsigned char *b = skb_tail_pointer(skb);
1191 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1193 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1194 tcm = NLMSG_DATA(nlh);
1195 tcm->tcm_family = AF_UNSPEC;
1196 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1197 tcm->tcm_parent = q->handle;
1198 tcm->tcm_handle = q->handle;
1200 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1201 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1202 goto nla_put_failure;
1204 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1205 TCA_XSTATS, &q->dev_queue->lock, &d) < 0)
1206 goto nla_put_failure;
1208 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1209 goto nla_put_failure;
1211 if (gnet_stats_finish_copy(&d) < 0)
1212 goto nla_put_failure;
1214 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1223 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1224 struct Qdisc *q, unsigned long cl, int event)
1226 struct sk_buff *skb;
1227 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1229 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1233 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1238 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1241 struct qdisc_dump_args
1243 struct qdisc_walker w;
1244 struct sk_buff *skb;
1245 struct netlink_callback *cb;
1248 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1250 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1252 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1253 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1256 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1258 struct net *net = sock_net(skb->sk);
1259 struct netdev_queue *dev_queue;
1262 struct net_device *dev;
1264 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1265 struct qdisc_dump_args arg;
1267 if (net != &init_net)
1270 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1272 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1278 dev_queue = &dev->tx_queue;
1279 list_for_each_entry(q, &dev_queue->qdisc_list, list) {
1280 if (t < s_t || !q->ops->cl_ops ||
1282 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1287 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1288 arg.w.fn = qdisc_class_dump;
1292 arg.w.skip = cb->args[1];
1294 q->ops->cl_ops->walk(q, &arg.w);
1295 cb->args[1] = arg.w.count;
1307 /* Main classifier routine: scans classifier chain attached
1308 to this qdisc, (optionally) tests for protocol and asks
1309 specific classifiers.
1311 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1312 struct tcf_result *res)
1314 __be16 protocol = skb->protocol;
1317 for (; tp; tp = tp->next) {
1318 if ((tp->protocol == protocol ||
1319 tp->protocol == htons(ETH_P_ALL)) &&
1320 (err = tp->classify(skb, tp, res)) >= 0) {
1321 #ifdef CONFIG_NET_CLS_ACT
1322 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1323 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1330 EXPORT_SYMBOL(tc_classify_compat);
1332 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1333 struct tcf_result *res)
1337 #ifdef CONFIG_NET_CLS_ACT
1338 struct tcf_proto *otp = tp;
1341 protocol = skb->protocol;
1343 err = tc_classify_compat(skb, tp, res);
1344 #ifdef CONFIG_NET_CLS_ACT
1345 if (err == TC_ACT_RECLASSIFY) {
1346 u32 verd = G_TC_VERD(skb->tc_verd);
1349 if (verd++ >= MAX_REC_LOOP) {
1350 printk("rule prio %u protocol %02x reclassify loop, "
1352 tp->prio&0xffff, ntohs(tp->protocol));
1355 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1361 EXPORT_SYMBOL(tc_classify);
1363 void tcf_destroy(struct tcf_proto *tp)
1365 tp->ops->destroy(tp);
1366 module_put(tp->ops->owner);
1370 void tcf_destroy_chain(struct tcf_proto **fl)
1372 struct tcf_proto *tp;
1374 while ((tp = *fl) != NULL) {
1379 EXPORT_SYMBOL(tcf_destroy_chain);
1381 #ifdef CONFIG_PROC_FS
1382 static int psched_show(struct seq_file *seq, void *v)
1386 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1387 seq_printf(seq, "%08x %08x %08x %08x\n",
1388 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1390 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1395 static int psched_open(struct inode *inode, struct file *file)
1397 return single_open(file, psched_show, PDE(inode)->data);
1400 static const struct file_operations psched_fops = {
1401 .owner = THIS_MODULE,
1402 .open = psched_open,
1404 .llseek = seq_lseek,
1405 .release = single_release,
1409 static int __init pktsched_init(void)
1411 register_qdisc(&pfifo_qdisc_ops);
1412 register_qdisc(&bfifo_qdisc_ops);
1413 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1415 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1416 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1417 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1418 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1419 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1420 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1425 subsys_initcall(pktsched_init);