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)
190 list_for_each_entry(q, &dev->qdisc_list, list) {
191 if (q->handle == handle)
197 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
201 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
205 cl = cops->get(p, classid);
209 leaf = cops->leaf(p, cl);
214 /* Find queueing discipline by name */
216 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
218 struct Qdisc_ops *q = NULL;
221 read_lock(&qdisc_mod_lock);
222 for (q = qdisc_base; q; q = q->next) {
223 if (nla_strcmp(kind, q->id) == 0) {
224 if (!try_module_get(q->owner))
229 read_unlock(&qdisc_mod_lock);
234 static struct qdisc_rate_table *qdisc_rtab_list;
236 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
238 struct qdisc_rate_table *rtab;
240 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
241 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
247 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
248 nla_len(tab) != TC_RTAB_SIZE)
251 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
255 memcpy(rtab->data, nla_data(tab), 1024);
256 rtab->next = qdisc_rtab_list;
257 qdisc_rtab_list = rtab;
261 EXPORT_SYMBOL(qdisc_get_rtab);
263 void qdisc_put_rtab(struct qdisc_rate_table *tab)
265 struct qdisc_rate_table *rtab, **rtabp;
267 if (!tab || --tab->refcnt)
270 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
278 EXPORT_SYMBOL(qdisc_put_rtab);
280 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
282 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
285 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
287 __netif_schedule(wd->qdisc);
289 return HRTIMER_NORESTART;
292 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
294 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
295 wd->timer.function = qdisc_watchdog;
298 EXPORT_SYMBOL(qdisc_watchdog_init);
300 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
304 wd->qdisc->flags |= TCQ_F_THROTTLED;
305 time = ktime_set(0, 0);
306 time = ktime_add_ns(time, PSCHED_US2NS(expires));
307 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
309 EXPORT_SYMBOL(qdisc_watchdog_schedule);
311 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
313 hrtimer_cancel(&wd->timer);
314 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
316 EXPORT_SYMBOL(qdisc_watchdog_cancel);
318 struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
320 unsigned int size = n * sizeof(struct hlist_head), i;
321 struct hlist_head *h;
323 if (size <= PAGE_SIZE)
324 h = kmalloc(size, GFP_KERNEL);
326 h = (struct hlist_head *)
327 __get_free_pages(GFP_KERNEL, get_order(size));
330 for (i = 0; i < n; i++)
331 INIT_HLIST_HEAD(&h[i]);
336 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
338 unsigned int size = n * sizeof(struct hlist_head);
340 if (size <= PAGE_SIZE)
343 free_pages((unsigned long)h, get_order(size));
346 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
348 struct Qdisc_class_common *cl;
349 struct hlist_node *n, *next;
350 struct hlist_head *nhash, *ohash;
351 unsigned int nsize, nmask, osize;
354 /* Rehash when load factor exceeds 0.75 */
355 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
357 nsize = clhash->hashsize * 2;
359 nhash = qdisc_class_hash_alloc(nsize);
363 ohash = clhash->hash;
364 osize = clhash->hashsize;
367 for (i = 0; i < osize; i++) {
368 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
369 h = qdisc_class_hash(cl->classid, nmask);
370 hlist_add_head(&cl->hnode, &nhash[h]);
373 clhash->hash = nhash;
374 clhash->hashsize = nsize;
375 clhash->hashmask = nmask;
376 sch_tree_unlock(sch);
378 qdisc_class_hash_free(ohash, osize);
380 EXPORT_SYMBOL(qdisc_class_hash_grow);
382 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
384 unsigned int size = 4;
386 clhash->hash = qdisc_class_hash_alloc(size);
387 if (clhash->hash == NULL)
389 clhash->hashsize = size;
390 clhash->hashmask = size - 1;
391 clhash->hashelems = 0;
394 EXPORT_SYMBOL(qdisc_class_hash_init);
396 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
398 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
400 EXPORT_SYMBOL(qdisc_class_hash_destroy);
402 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
403 struct Qdisc_class_common *cl)
407 INIT_HLIST_NODE(&cl->hnode);
408 h = qdisc_class_hash(cl->classid, clhash->hashmask);
409 hlist_add_head(&cl->hnode, &clhash->hash[h]);
412 EXPORT_SYMBOL(qdisc_class_hash_insert);
414 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
415 struct Qdisc_class_common *cl)
417 hlist_del(&cl->hnode);
420 EXPORT_SYMBOL(qdisc_class_hash_remove);
422 /* Allocate an unique handle from space managed by kernel */
424 static u32 qdisc_alloc_handle(struct net_device *dev)
427 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
430 autohandle += TC_H_MAKE(0x10000U, 0);
431 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
432 autohandle = TC_H_MAKE(0x80000000U, 0);
433 } while (qdisc_lookup(dev, autohandle) && --i > 0);
435 return i>0 ? autohandle : 0;
438 /* Attach toplevel qdisc to device dev */
440 static struct Qdisc *
441 dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc)
443 struct netdev_queue *dev_queue;
444 struct Qdisc *oqdisc;
446 if (dev->flags & IFF_UP)
449 qdisc_lock_tree(dev);
450 if (qdisc && qdisc->flags&TCQ_F_INGRESS) {
451 dev_queue = &dev->rx_queue;
452 oqdisc = dev_queue->qdisc;
453 /* Prune old scheduler */
454 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) {
457 dev_queue->qdisc = NULL;
459 dev_queue->qdisc = qdisc;
463 dev_queue = netdev_get_tx_queue(dev, 0);
464 oqdisc = dev_queue->qdisc_sleeping;
466 /* Prune old scheduler */
467 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
470 /* ... and graft new one */
473 dev_queue->qdisc_sleeping = qdisc;
474 dev_queue->qdisc = &noop_qdisc;
477 qdisc_unlock_tree(dev);
479 if (dev->flags & IFF_UP)
485 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
487 const struct Qdisc_class_ops *cops;
493 while ((parentid = sch->parent)) {
494 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
497 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
499 WARN_ON(parentid != TC_H_ROOT);
502 cops = sch->ops->cl_ops;
503 if (cops->qlen_notify) {
504 cl = cops->get(sch, parentid);
505 cops->qlen_notify(sch, cl);
511 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
513 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
516 Old qdisc is not destroyed but returned in *old.
519 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
521 struct Qdisc *new, struct Qdisc **old)
524 struct Qdisc *q = *old;
527 if (parent == NULL) {
528 if (q && q->flags&TCQ_F_INGRESS) {
529 *old = dev_graft_qdisc(dev, q);
531 *old = dev_graft_qdisc(dev, new);
534 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
539 unsigned long cl = cops->get(parent, classid);
541 err = cops->graft(parent, cl, new, old);
542 cops->put(parent, cl);
550 Allocate and initialize new qdisc.
552 Parameters are passed via opt.
555 static struct Qdisc *
556 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
557 u32 parent, u32 handle, struct nlattr **tca, int *errp)
560 struct nlattr *kind = tca[TCA_KIND];
562 struct Qdisc_ops *ops;
564 ops = qdisc_lookup_ops(kind);
566 if (ops == NULL && kind != NULL) {
568 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
569 /* We dropped the RTNL semaphore in order to
570 * perform the module load. So, even if we
571 * succeeded in loading the module we have to
572 * tell the caller to replay the request. We
573 * indicate this using -EAGAIN.
574 * We replay the request because the device may
575 * go away in the mean time.
578 request_module("sch_%s", name);
580 ops = qdisc_lookup_ops(kind);
582 /* We will try again qdisc_lookup_ops,
583 * so don't keep a reference.
585 module_put(ops->owner);
597 sch = qdisc_alloc(dev_queue, ops);
603 sch->parent = parent;
605 if (handle == TC_H_INGRESS) {
606 sch->flags |= TCQ_F_INGRESS;
607 handle = TC_H_MAKE(TC_H_INGRESS, 0);
610 handle = qdisc_alloc_handle(dev);
617 sch->handle = handle;
619 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
621 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
622 qdisc_root_lock(sch),
626 * Any broken qdiscs that would require
627 * a ops->reset() here? The qdisc was never
628 * in action so it shouldn't be necessary.
635 spin_lock_bh(&dev->qdisc_list_lock);
636 list_add_tail(&sch->list, &dev->qdisc_list);
637 spin_unlock_bh(&dev->qdisc_list_lock);
643 kfree((char *) sch - sch->padded);
645 module_put(ops->owner);
651 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
653 if (tca[TCA_OPTIONS]) {
656 if (sch->ops->change == NULL)
658 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
663 gen_replace_estimator(&sch->bstats, &sch->rate_est,
664 qdisc_root_lock(sch), tca[TCA_RATE]);
668 struct check_loop_arg
670 struct qdisc_walker w;
675 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
677 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
679 struct check_loop_arg arg;
681 if (q->ops->cl_ops == NULL)
684 arg.w.stop = arg.w.skip = arg.w.count = 0;
685 arg.w.fn = check_loop_fn;
688 q->ops->cl_ops->walk(q, &arg.w);
689 return arg.w.stop ? -ELOOP : 0;
693 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
696 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
697 struct check_loop_arg *arg = (struct check_loop_arg *)w;
699 leaf = cops->leaf(q, cl);
701 if (leaf == arg->p || arg->depth > 7)
703 return check_loop(leaf, arg->p, arg->depth + 1);
712 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
714 struct net *net = sock_net(skb->sk);
715 struct tcmsg *tcm = NLMSG_DATA(n);
716 struct nlattr *tca[TCA_MAX + 1];
717 struct net_device *dev;
718 u32 clid = tcm->tcm_parent;
719 struct Qdisc *q = NULL;
720 struct Qdisc *p = NULL;
723 if (net != &init_net)
726 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
729 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
734 if (clid != TC_H_ROOT) {
735 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
736 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
738 q = qdisc_leaf(p, clid);
739 } else { /* ingress */
740 q = dev->rx_queue.qdisc;
743 struct netdev_queue *dev_queue;
744 dev_queue = netdev_get_tx_queue(dev, 0);
745 q = dev_queue->qdisc_sleeping;
750 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
753 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
757 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
760 if (n->nlmsg_type == RTM_DELQDISC) {
765 if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0)
768 qdisc_notify(skb, n, clid, q, NULL);
769 qdisc_lock_tree(dev);
771 qdisc_unlock_tree(dev);
774 qdisc_notify(skb, n, clid, NULL, q);
783 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
785 struct net *net = sock_net(skb->sk);
787 struct nlattr *tca[TCA_MAX + 1];
788 struct net_device *dev;
793 if (net != &init_net)
797 /* Reinit, just in case something touches this. */
799 clid = tcm->tcm_parent;
802 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
805 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
810 if (clid != TC_H_ROOT) {
811 if (clid != TC_H_INGRESS) {
812 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
814 q = qdisc_leaf(p, clid);
815 } else { /*ingress */
816 q = dev->rx_queue.qdisc;
819 struct netdev_queue *dev_queue;
820 dev_queue = netdev_get_tx_queue(dev, 0);
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, netdev_get_tx_queue(dev, 0),
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, qdisc_root_lock(q), &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) {
1027 list_for_each_entry(q, &dev->qdisc_list, list) {
1028 if (q_idx < s_q_idx) {
1032 if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1033 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1042 read_unlock(&dev_base_lock);
1045 cb->args[1] = q_idx;
1052 /************************************************
1053 * Traffic classes manipulation. *
1054 ************************************************/
1058 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1060 struct net *net = sock_net(skb->sk);
1061 struct netdev_queue *dev_queue;
1062 struct tcmsg *tcm = NLMSG_DATA(n);
1063 struct nlattr *tca[TCA_MAX + 1];
1064 struct net_device *dev;
1065 struct Qdisc *q = NULL;
1066 const struct Qdisc_class_ops *cops;
1067 unsigned long cl = 0;
1068 unsigned long new_cl;
1069 u32 pid = tcm->tcm_parent;
1070 u32 clid = tcm->tcm_handle;
1071 u32 qid = TC_H_MAJ(clid);
1074 if (net != &init_net)
1077 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1080 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1085 parent == TC_H_UNSPEC - unspecified parent.
1086 parent == TC_H_ROOT - class is root, which has no parent.
1087 parent == X:0 - parent is root class.
1088 parent == X:Y - parent is a node in hierarchy.
1089 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1091 handle == 0:0 - generate handle from kernel pool.
1092 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1093 handle == X:Y - clear.
1094 handle == X:0 - root class.
1097 /* Step 1. Determine qdisc handle X:0 */
1099 dev_queue = netdev_get_tx_queue(dev, 0);
1100 if (pid != TC_H_ROOT) {
1101 u32 qid1 = TC_H_MAJ(pid);
1104 /* If both majors are known, they must be identical. */
1109 } else if (qid == 0)
1110 qid = dev_queue->qdisc_sleeping->handle;
1112 /* Now qid is genuine qdisc handle consistent
1113 both with parent and child.
1115 TC_H_MAJ(pid) still may be unspecified, complete it now.
1118 pid = TC_H_MAKE(qid, pid);
1121 qid = dev_queue->qdisc_sleeping->handle;
1124 /* OK. Locate qdisc */
1125 if ((q = qdisc_lookup(dev, qid)) == NULL)
1128 /* An check that it supports classes */
1129 cops = q->ops->cl_ops;
1133 /* Now try to get class */
1135 if (pid == TC_H_ROOT)
1138 clid = TC_H_MAKE(qid, clid);
1141 cl = cops->get(q, clid);
1145 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1148 switch (n->nlmsg_type) {
1151 if (n->nlmsg_flags&NLM_F_EXCL)
1155 err = cops->delete(q, cl);
1157 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1160 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1169 err = cops->change(q, clid, pid, tca, &new_cl);
1171 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1181 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1183 u32 pid, u32 seq, u16 flags, int event)
1186 struct nlmsghdr *nlh;
1187 unsigned char *b = skb_tail_pointer(skb);
1189 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1191 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1192 tcm = NLMSG_DATA(nlh);
1193 tcm->tcm_family = AF_UNSPEC;
1194 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1195 tcm->tcm_parent = q->handle;
1196 tcm->tcm_handle = q->handle;
1198 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1199 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1200 goto nla_put_failure;
1202 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1203 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
1204 goto nla_put_failure;
1206 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1207 goto nla_put_failure;
1209 if (gnet_stats_finish_copy(&d) < 0)
1210 goto nla_put_failure;
1212 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1221 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1222 struct Qdisc *q, unsigned long cl, int event)
1224 struct sk_buff *skb;
1225 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1227 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1231 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1236 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1239 struct qdisc_dump_args
1241 struct qdisc_walker w;
1242 struct sk_buff *skb;
1243 struct netlink_callback *cb;
1246 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1248 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1250 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1251 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1254 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1256 struct net *net = sock_net(skb->sk);
1259 struct net_device *dev;
1261 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1262 struct qdisc_dump_args arg;
1264 if (net != &init_net)
1267 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1269 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1275 list_for_each_entry(q, &dev->qdisc_list, list) {
1276 if (t < s_t || !q->ops->cl_ops ||
1278 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1283 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1284 arg.w.fn = qdisc_class_dump;
1288 arg.w.skip = cb->args[1];
1290 q->ops->cl_ops->walk(q, &arg.w);
1291 cb->args[1] = arg.w.count;
1303 /* Main classifier routine: scans classifier chain attached
1304 to this qdisc, (optionally) tests for protocol and asks
1305 specific classifiers.
1307 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1308 struct tcf_result *res)
1310 __be16 protocol = skb->protocol;
1313 for (; tp; tp = tp->next) {
1314 if ((tp->protocol == protocol ||
1315 tp->protocol == htons(ETH_P_ALL)) &&
1316 (err = tp->classify(skb, tp, res)) >= 0) {
1317 #ifdef CONFIG_NET_CLS_ACT
1318 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1319 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1326 EXPORT_SYMBOL(tc_classify_compat);
1328 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1329 struct tcf_result *res)
1333 #ifdef CONFIG_NET_CLS_ACT
1334 struct tcf_proto *otp = tp;
1337 protocol = skb->protocol;
1339 err = tc_classify_compat(skb, tp, res);
1340 #ifdef CONFIG_NET_CLS_ACT
1341 if (err == TC_ACT_RECLASSIFY) {
1342 u32 verd = G_TC_VERD(skb->tc_verd);
1345 if (verd++ >= MAX_REC_LOOP) {
1346 printk("rule prio %u protocol %02x reclassify loop, "
1348 tp->prio&0xffff, ntohs(tp->protocol));
1351 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1357 EXPORT_SYMBOL(tc_classify);
1359 void tcf_destroy(struct tcf_proto *tp)
1361 tp->ops->destroy(tp);
1362 module_put(tp->ops->owner);
1366 void tcf_destroy_chain(struct tcf_proto **fl)
1368 struct tcf_proto *tp;
1370 while ((tp = *fl) != NULL) {
1375 EXPORT_SYMBOL(tcf_destroy_chain);
1377 #ifdef CONFIG_PROC_FS
1378 static int psched_show(struct seq_file *seq, void *v)
1382 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1383 seq_printf(seq, "%08x %08x %08x %08x\n",
1384 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1386 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1391 static int psched_open(struct inode *inode, struct file *file)
1393 return single_open(file, psched_show, PDE(inode)->data);
1396 static const struct file_operations psched_fops = {
1397 .owner = THIS_MODULE,
1398 .open = psched_open,
1400 .llseek = seq_lseek,
1401 .release = single_release,
1405 static int __init pktsched_init(void)
1407 register_qdisc(&pfifo_qdisc_ops);
1408 register_qdisc(&bfifo_qdisc_ops);
1409 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1411 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1412 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1413 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1414 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1415 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1416 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1421 subsys_initcall(pktsched_init);