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/netlink.h>
32 #include <net/pkt_sched.h>
34 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid,
35 struct Qdisc *old, struct Qdisc *new);
36 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
37 struct Qdisc *q, unsigned long cl, int event);
44 This file consists of two interrelated parts:
46 1. queueing disciplines manager frontend.
47 2. traffic classes manager frontend.
49 Generally, queueing discipline ("qdisc") is a black box,
50 which is able to enqueue packets and to dequeue them (when
51 device is ready to send something) in order and at times
52 determined by algorithm hidden in it.
54 qdisc's are divided to two categories:
55 - "queues", which have no internal structure visible from outside.
56 - "schedulers", which split all the packets to "traffic classes",
57 using "packet classifiers" (look at cls_api.c)
59 In turn, classes may have child qdiscs (as rule, queues)
60 attached to them etc. etc. etc.
62 The goal of the routines in this file is to translate
63 information supplied by user in the form of handles
64 to more intelligible for kernel form, to make some sanity
65 checks and part of work, which is common to all qdiscs
66 and to provide rtnetlink notifications.
68 All real intelligent work is done inside qdisc modules.
72 Every discipline has two major routines: enqueue and dequeue.
76 dequeue usually returns a skb to send. It is allowed to return NULL,
77 but it does not mean that queue is empty, it just means that
78 discipline does not want to send anything this time.
79 Queue is really empty if q->q.qlen == 0.
80 For complicated disciplines with multiple queues q->q is not
81 real packet queue, but however q->q.qlen must be valid.
85 enqueue returns 0, if packet was enqueued successfully.
86 If packet (this one or another one) was dropped, it returns
88 NET_XMIT_DROP - this packet dropped
89 Expected action: do not backoff, but wait until queue will clear.
90 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
91 Expected action: backoff or ignore
92 NET_XMIT_POLICED - dropped by police.
93 Expected action: backoff or error to real-time apps.
99 requeues once dequeued packet. It is used for non-standard or
100 just buggy devices, which can defer output even if dev->tbusy=0.
104 returns qdisc to initial state: purge all buffers, clear all
105 timers, counters (except for statistics) etc.
109 initializes newly created qdisc.
113 destroys resources allocated by init and during lifetime of qdisc.
117 changes qdisc parameters.
120 /* Protects list of registered TC modules. It is pure SMP lock. */
121 static DEFINE_RWLOCK(qdisc_mod_lock);
124 /************************************************
125 * Queueing disciplines manipulation. *
126 ************************************************/
129 /* The list of all installed queueing disciplines. */
131 static struct Qdisc_ops *qdisc_base;
133 /* Register/uregister queueing discipline */
135 int register_qdisc(struct Qdisc_ops *qops)
137 struct Qdisc_ops *q, **qp;
140 write_lock(&qdisc_mod_lock);
141 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
142 if (!strcmp(qops->id, q->id))
145 if (qops->enqueue == NULL)
146 qops->enqueue = noop_qdisc_ops.enqueue;
147 if (qops->requeue == NULL)
148 qops->requeue = noop_qdisc_ops.requeue;
149 if (qops->dequeue == NULL)
150 qops->dequeue = noop_qdisc_ops.dequeue;
156 write_unlock(&qdisc_mod_lock);
160 int unregister_qdisc(struct Qdisc_ops *qops)
162 struct Qdisc_ops *q, **qp;
165 write_lock(&qdisc_mod_lock);
166 for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next)
174 write_unlock(&qdisc_mod_lock);
178 /* We know handle. Find qdisc among all qdisc's attached to device
179 (root qdisc, all its children, children of children etc.)
182 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
186 list_for_each_entry(q, &dev->qdisc_list, list) {
187 if (q->handle == handle)
193 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
197 struct Qdisc_class_ops *cops = p->ops->cl_ops;
201 cl = cops->get(p, classid);
205 leaf = cops->leaf(p, cl);
210 /* Find queueing discipline by name */
212 static struct Qdisc_ops *qdisc_lookup_ops(struct rtattr *kind)
214 struct Qdisc_ops *q = NULL;
217 read_lock(&qdisc_mod_lock);
218 for (q = qdisc_base; q; q = q->next) {
219 if (rtattr_strcmp(kind, q->id) == 0) {
220 if (!try_module_get(q->owner))
225 read_unlock(&qdisc_mod_lock);
230 static struct qdisc_rate_table *qdisc_rtab_list;
232 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct rtattr *tab)
234 struct qdisc_rate_table *rtab;
236 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
237 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
243 if (tab == NULL || r->rate == 0 || r->cell_log == 0 || RTA_PAYLOAD(tab) != 1024)
246 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
250 memcpy(rtab->data, RTA_DATA(tab), 1024);
251 rtab->next = qdisc_rtab_list;
252 qdisc_rtab_list = rtab;
257 void qdisc_put_rtab(struct qdisc_rate_table *tab)
259 struct qdisc_rate_table *rtab, **rtabp;
261 if (!tab || --tab->refcnt)
264 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
273 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
275 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
277 struct net_device *dev = wd->qdisc->dev;
279 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
283 return HRTIMER_NORESTART;
286 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
288 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
289 wd->timer.function = qdisc_watchdog;
292 EXPORT_SYMBOL(qdisc_watchdog_init);
294 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
298 wd->qdisc->flags |= TCQ_F_THROTTLED;
299 time = ktime_set(0, 0);
300 time = ktime_add_ns(time, PSCHED_US2NS(expires));
301 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
303 EXPORT_SYMBOL(qdisc_watchdog_schedule);
305 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
307 hrtimer_cancel(&wd->timer);
308 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
310 EXPORT_SYMBOL(qdisc_watchdog_cancel);
312 /* Allocate an unique handle from space managed by kernel */
314 static u32 qdisc_alloc_handle(struct net_device *dev)
317 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
320 autohandle += TC_H_MAKE(0x10000U, 0);
321 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
322 autohandle = TC_H_MAKE(0x80000000U, 0);
323 } while (qdisc_lookup(dev, autohandle) && --i > 0);
325 return i>0 ? autohandle : 0;
328 /* Attach toplevel qdisc to device dev */
330 static struct Qdisc *
331 dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc)
333 struct Qdisc *oqdisc;
335 if (dev->flags & IFF_UP)
338 qdisc_lock_tree(dev);
339 if (qdisc && qdisc->flags&TCQ_F_INGRESS) {
340 oqdisc = dev->qdisc_ingress;
341 /* Prune old scheduler */
342 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) {
345 dev->qdisc_ingress = NULL;
347 dev->qdisc_ingress = qdisc;
352 oqdisc = dev->qdisc_sleeping;
354 /* Prune old scheduler */
355 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
358 /* ... and graft new one */
361 dev->qdisc_sleeping = qdisc;
362 dev->qdisc = &noop_qdisc;
365 qdisc_unlock_tree(dev);
367 if (dev->flags & IFF_UP)
373 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
375 struct Qdisc_class_ops *cops;
381 while ((parentid = sch->parent)) {
382 sch = qdisc_lookup(sch->dev, TC_H_MAJ(parentid));
384 WARN_ON(parentid != TC_H_ROOT);
387 cops = sch->ops->cl_ops;
388 if (cops->qlen_notify) {
389 cl = cops->get(sch, parentid);
390 cops->qlen_notify(sch, cl);
396 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
398 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
401 Old qdisc is not destroyed but returned in *old.
404 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
406 struct Qdisc *new, struct Qdisc **old)
409 struct Qdisc *q = *old;
412 if (parent == NULL) {
413 if (q && q->flags&TCQ_F_INGRESS) {
414 *old = dev_graft_qdisc(dev, q);
416 *old = dev_graft_qdisc(dev, new);
419 struct Qdisc_class_ops *cops = parent->ops->cl_ops;
424 unsigned long cl = cops->get(parent, classid);
426 err = cops->graft(parent, cl, new, old);
427 cops->put(parent, cl);
435 Allocate and initialize new qdisc.
437 Parameters are passed via opt.
440 static struct Qdisc *
441 qdisc_create(struct net_device *dev, u32 parent, u32 handle,
442 struct rtattr **tca, int *errp)
445 struct rtattr *kind = tca[TCA_KIND-1];
447 struct Qdisc_ops *ops;
449 ops = qdisc_lookup_ops(kind);
451 if (ops == NULL && kind != NULL) {
453 if (rtattr_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
454 /* We dropped the RTNL semaphore in order to
455 * perform the module load. So, even if we
456 * succeeded in loading the module we have to
457 * tell the caller to replay the request. We
458 * indicate this using -EAGAIN.
459 * We replay the request because the device may
460 * go away in the mean time.
463 request_module("sch_%s", name);
465 ops = qdisc_lookup_ops(kind);
467 /* We will try again qdisc_lookup_ops,
468 * so don't keep a reference.
470 module_put(ops->owner);
482 sch = qdisc_alloc(dev, ops);
488 sch->parent = parent;
490 if (handle == TC_H_INGRESS) {
491 sch->flags |= TCQ_F_INGRESS;
492 sch->stats_lock = &dev->ingress_lock;
493 handle = TC_H_MAKE(TC_H_INGRESS, 0);
495 sch->stats_lock = &dev->queue_lock;
497 handle = qdisc_alloc_handle(dev);
504 sch->handle = handle;
506 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS-1])) == 0) {
507 if (tca[TCA_RATE-1]) {
508 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
513 * Any broken qdiscs that would require
514 * a ops->reset() here? The qdisc was never
515 * in action so it shouldn't be necessary.
522 qdisc_lock_tree(dev);
523 list_add_tail(&sch->list, &dev->qdisc_list);
524 qdisc_unlock_tree(dev);
530 kfree((char *) sch - sch->padded);
532 module_put(ops->owner);
538 static int qdisc_change(struct Qdisc *sch, struct rtattr **tca)
540 if (tca[TCA_OPTIONS-1]) {
543 if (sch->ops->change == NULL)
545 err = sch->ops->change(sch, tca[TCA_OPTIONS-1]);
550 gen_replace_estimator(&sch->bstats, &sch->rate_est,
551 sch->stats_lock, tca[TCA_RATE-1]);
555 struct check_loop_arg
557 struct qdisc_walker w;
562 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
564 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
566 struct check_loop_arg arg;
568 if (q->ops->cl_ops == NULL)
571 arg.w.stop = arg.w.skip = arg.w.count = 0;
572 arg.w.fn = check_loop_fn;
575 q->ops->cl_ops->walk(q, &arg.w);
576 return arg.w.stop ? -ELOOP : 0;
580 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
583 struct Qdisc_class_ops *cops = q->ops->cl_ops;
584 struct check_loop_arg *arg = (struct check_loop_arg *)w;
586 leaf = cops->leaf(q, cl);
588 if (leaf == arg->p || arg->depth > 7)
590 return check_loop(leaf, arg->p, arg->depth + 1);
599 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
601 struct tcmsg *tcm = NLMSG_DATA(n);
602 struct rtattr **tca = arg;
603 struct net_device *dev;
604 u32 clid = tcm->tcm_parent;
605 struct Qdisc *q = NULL;
606 struct Qdisc *p = NULL;
609 if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL)
613 if (clid != TC_H_ROOT) {
614 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
615 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
617 q = qdisc_leaf(p, clid);
618 } else { /* ingress */
619 q = dev->qdisc_ingress;
622 q = dev->qdisc_sleeping;
627 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
630 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
634 if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id))
637 if (n->nlmsg_type == RTM_DELQDISC) {
642 if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0)
645 qdisc_notify(skb, n, clid, q, NULL);
646 qdisc_lock_tree(dev);
648 qdisc_unlock_tree(dev);
651 qdisc_notify(skb, n, clid, NULL, q);
660 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
664 struct net_device *dev;
670 /* Reinit, just in case something touches this. */
673 clid = tcm->tcm_parent;
676 if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL)
680 if (clid != TC_H_ROOT) {
681 if (clid != TC_H_INGRESS) {
682 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
684 q = qdisc_leaf(p, clid);
685 } else { /*ingress */
686 q = dev->qdisc_ingress;
689 q = dev->qdisc_sleeping;
692 /* It may be default qdisc, ignore it */
693 if (q && q->handle == 0)
696 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
697 if (tcm->tcm_handle) {
698 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
700 if (TC_H_MIN(tcm->tcm_handle))
702 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
704 if (n->nlmsg_flags&NLM_F_EXCL)
706 if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id))
709 (p && check_loop(q, p, 0)))
711 atomic_inc(&q->refcnt);
717 /* This magic test requires explanation.
719 * We know, that some child q is already
720 * attached to this parent and have choice:
721 * either to change it or to create/graft new one.
723 * 1. We are allowed to create/graft only
724 * if CREATE and REPLACE flags are set.
726 * 2. If EXCL is set, requestor wanted to say,
727 * that qdisc tcm_handle is not expected
728 * to exist, so that we choose create/graft too.
730 * 3. The last case is when no flags are set.
731 * Alas, it is sort of hole in API, we
732 * cannot decide what to do unambiguously.
733 * For now we select create/graft, if
734 * user gave KIND, which does not match existing.
736 if ((n->nlmsg_flags&NLM_F_CREATE) &&
737 (n->nlmsg_flags&NLM_F_REPLACE) &&
738 ((n->nlmsg_flags&NLM_F_EXCL) ||
740 rtattr_strcmp(tca[TCA_KIND-1], q->ops->id))))
745 if (!tcm->tcm_handle)
747 q = qdisc_lookup(dev, tcm->tcm_handle);
750 /* Change qdisc parameters */
753 if (n->nlmsg_flags&NLM_F_EXCL)
755 if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], q->ops->id))
757 err = qdisc_change(q, tca);
759 qdisc_notify(skb, n, clid, NULL, q);
763 if (!(n->nlmsg_flags&NLM_F_CREATE))
765 if (clid == TC_H_INGRESS)
766 q = qdisc_create(dev, tcm->tcm_parent, tcm->tcm_parent,
769 q = qdisc_create(dev, tcm->tcm_parent, tcm->tcm_handle,
779 struct Qdisc *old_q = NULL;
780 err = qdisc_graft(dev, p, clid, q, &old_q);
783 qdisc_lock_tree(dev);
785 qdisc_unlock_tree(dev);
789 qdisc_notify(skb, n, clid, old_q, q);
791 qdisc_lock_tree(dev);
792 qdisc_destroy(old_q);
793 qdisc_unlock_tree(dev);
799 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
800 u32 pid, u32 seq, u16 flags, int event)
803 struct nlmsghdr *nlh;
804 unsigned char *b = skb_tail_pointer(skb);
807 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
808 tcm = NLMSG_DATA(nlh);
809 tcm->tcm_family = AF_UNSPEC;
812 tcm->tcm_ifindex = q->dev->ifindex;
813 tcm->tcm_parent = clid;
814 tcm->tcm_handle = q->handle;
815 tcm->tcm_info = atomic_read(&q->refcnt);
816 RTA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id);
817 if (q->ops->dump && q->ops->dump(q, skb) < 0)
819 q->qstats.qlen = q->q.qlen;
821 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
822 TCA_XSTATS, q->stats_lock, &d) < 0)
825 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
828 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
829 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
830 gnet_stats_copy_queue(&d, &q->qstats) < 0)
833 if (gnet_stats_finish_copy(&d) < 0)
836 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
845 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
846 u32 clid, struct Qdisc *old, struct Qdisc *new)
849 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
851 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
855 if (old && old->handle) {
856 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
860 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
865 return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
872 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
876 struct net_device *dev;
880 s_q_idx = q_idx = cb->args[1];
881 read_lock(&dev_base_lock);
883 for_each_netdev(dev) {
889 list_for_each_entry(q, &dev->qdisc_list, list) {
890 if (q_idx < s_q_idx) {
894 if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
895 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
904 read_unlock(&dev_base_lock);
914 /************************************************
915 * Traffic classes manipulation. *
916 ************************************************/
920 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
922 struct tcmsg *tcm = NLMSG_DATA(n);
923 struct rtattr **tca = arg;
924 struct net_device *dev;
925 struct Qdisc *q = NULL;
926 struct Qdisc_class_ops *cops;
927 unsigned long cl = 0;
928 unsigned long new_cl;
929 u32 pid = tcm->tcm_parent;
930 u32 clid = tcm->tcm_handle;
931 u32 qid = TC_H_MAJ(clid);
934 if ((dev = __dev_get_by_index(tcm->tcm_ifindex)) == NULL)
938 parent == TC_H_UNSPEC - unspecified parent.
939 parent == TC_H_ROOT - class is root, which has no parent.
940 parent == X:0 - parent is root class.
941 parent == X:Y - parent is a node in hierarchy.
942 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
944 handle == 0:0 - generate handle from kernel pool.
945 handle == 0:Y - class is X:Y, where X:0 is qdisc.
946 handle == X:Y - clear.
947 handle == X:0 - root class.
950 /* Step 1. Determine qdisc handle X:0 */
952 if (pid != TC_H_ROOT) {
953 u32 qid1 = TC_H_MAJ(pid);
956 /* If both majors are known, they must be identical. */
962 qid = dev->qdisc_sleeping->handle;
964 /* Now qid is genuine qdisc handle consistent
965 both with parent and child.
967 TC_H_MAJ(pid) still may be unspecified, complete it now.
970 pid = TC_H_MAKE(qid, pid);
973 qid = dev->qdisc_sleeping->handle;
976 /* OK. Locate qdisc */
977 if ((q = qdisc_lookup(dev, qid)) == NULL)
980 /* An check that it supports classes */
981 cops = q->ops->cl_ops;
985 /* Now try to get class */
987 if (pid == TC_H_ROOT)
990 clid = TC_H_MAKE(qid, clid);
993 cl = cops->get(q, clid);
997 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1000 switch (n->nlmsg_type) {
1003 if (n->nlmsg_flags&NLM_F_EXCL)
1007 err = cops->delete(q, cl);
1009 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1012 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1021 err = cops->change(q, clid, pid, tca, &new_cl);
1023 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1033 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1035 u32 pid, u32 seq, u16 flags, int event)
1038 struct nlmsghdr *nlh;
1039 unsigned char *b = skb_tail_pointer(skb);
1041 struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1043 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1044 tcm = NLMSG_DATA(nlh);
1045 tcm->tcm_family = AF_UNSPEC;
1046 tcm->tcm_ifindex = q->dev->ifindex;
1047 tcm->tcm_parent = q->handle;
1048 tcm->tcm_handle = q->handle;
1050 RTA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id);
1051 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1052 goto rtattr_failure;
1054 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1055 TCA_XSTATS, q->stats_lock, &d) < 0)
1056 goto rtattr_failure;
1058 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1059 goto rtattr_failure;
1061 if (gnet_stats_finish_copy(&d) < 0)
1062 goto rtattr_failure;
1064 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1073 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1074 struct Qdisc *q, unsigned long cl, int event)
1076 struct sk_buff *skb;
1077 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1079 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1083 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1088 return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1091 struct qdisc_dump_args
1093 struct qdisc_walker w;
1094 struct sk_buff *skb;
1095 struct netlink_callback *cb;
1098 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1100 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1102 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1103 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1106 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1110 struct net_device *dev;
1112 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1113 struct qdisc_dump_args arg;
1115 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1117 if ((dev = dev_get_by_index(tcm->tcm_ifindex)) == NULL)
1123 list_for_each_entry(q, &dev->qdisc_list, list) {
1124 if (t < s_t || !q->ops->cl_ops ||
1126 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1131 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1132 arg.w.fn = qdisc_class_dump;
1136 arg.w.skip = cb->args[1];
1138 q->ops->cl_ops->walk(q, &arg.w);
1139 cb->args[1] = arg.w.count;
1151 /* Main classifier routine: scans classifier chain attached
1152 to this qdisc, (optionally) tests for protocol and asks
1153 specific classifiers.
1155 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1156 struct tcf_result *res)
1158 __be16 protocol = skb->protocol;
1161 for (; tp; tp = tp->next) {
1162 if ((tp->protocol == protocol ||
1163 tp->protocol == htons(ETH_P_ALL)) &&
1164 (err = tp->classify(skb, tp, res)) >= 0) {
1165 #ifdef CONFIG_NET_CLS_ACT
1166 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1167 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1174 EXPORT_SYMBOL(tc_classify_compat);
1176 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1177 struct tcf_result *res)
1181 #ifdef CONFIG_NET_CLS_ACT
1182 struct tcf_proto *otp = tp;
1185 protocol = skb->protocol;
1187 err = tc_classify_compat(skb, tp, res);
1188 #ifdef CONFIG_NET_CLS_ACT
1189 if (err == TC_ACT_RECLASSIFY) {
1190 u32 verd = G_TC_VERD(skb->tc_verd);
1193 if (verd++ >= MAX_REC_LOOP) {
1194 printk("rule prio %u protocol %02x reclassify loop, "
1196 tp->prio&0xffff, ntohs(tp->protocol));
1199 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1205 EXPORT_SYMBOL(tc_classify);
1207 void tcf_destroy(struct tcf_proto *tp)
1209 tp->ops->destroy(tp);
1210 module_put(tp->ops->owner);
1214 void tcf_destroy_chain(struct tcf_proto *fl)
1216 struct tcf_proto *tp;
1218 while ((tp = fl) != NULL) {
1223 EXPORT_SYMBOL(tcf_destroy_chain);
1225 #ifdef CONFIG_PROC_FS
1226 static int psched_show(struct seq_file *seq, void *v)
1228 seq_printf(seq, "%08x %08x %08x %08x\n",
1229 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1231 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(KTIME_MONOTONIC_RES));
1236 static int psched_open(struct inode *inode, struct file *file)
1238 return single_open(file, psched_show, PDE(inode)->data);
1241 static const struct file_operations psched_fops = {
1242 .owner = THIS_MODULE,
1243 .open = psched_open,
1245 .llseek = seq_lseek,
1246 .release = single_release,
1250 static int __init pktsched_init(void)
1252 register_qdisc(&pfifo_qdisc_ops);
1253 register_qdisc(&bfifo_qdisc_ops);
1254 proc_net_fops_create("psched", 0, &psched_fops);
1256 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1257 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1258 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1259 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1260 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1261 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1266 subsys_initcall(pktsched_init);
1268 EXPORT_SYMBOL(qdisc_get_rtab);
1269 EXPORT_SYMBOL(qdisc_put_rtab);
1270 EXPORT_SYMBOL(register_qdisc);
1271 EXPORT_SYMBOL(unregister_qdisc);