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 dev->tbusy=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 || nla_len(tab) != 1024)
250 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
254 memcpy(rtab->data, nla_data(tab), 1024);
255 rtab->next = qdisc_rtab_list;
256 qdisc_rtab_list = rtab;
260 EXPORT_SYMBOL(qdisc_get_rtab);
262 void qdisc_put_rtab(struct qdisc_rate_table *tab)
264 struct qdisc_rate_table *rtab, **rtabp;
266 if (!tab || --tab->refcnt)
269 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
277 EXPORT_SYMBOL(qdisc_put_rtab);
279 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
281 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
283 struct net_device *dev = wd->qdisc->dev;
285 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
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 /* Allocate an unique handle from space managed by kernel */
320 static u32 qdisc_alloc_handle(struct net_device *dev)
323 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
326 autohandle += TC_H_MAKE(0x10000U, 0);
327 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
328 autohandle = TC_H_MAKE(0x80000000U, 0);
329 } while (qdisc_lookup(dev, autohandle) && --i > 0);
331 return i>0 ? autohandle : 0;
334 /* Attach toplevel qdisc to device dev */
336 static struct Qdisc *
337 dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc)
339 struct Qdisc *oqdisc;
341 if (dev->flags & IFF_UP)
344 qdisc_lock_tree(dev);
345 if (qdisc && qdisc->flags&TCQ_F_INGRESS) {
346 oqdisc = dev->qdisc_ingress;
347 /* Prune old scheduler */
348 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) {
351 dev->qdisc_ingress = NULL;
353 dev->qdisc_ingress = qdisc;
358 oqdisc = dev->qdisc_sleeping;
360 /* Prune old scheduler */
361 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
364 /* ... and graft new one */
367 dev->qdisc_sleeping = qdisc;
368 dev->qdisc = &noop_qdisc;
371 qdisc_unlock_tree(dev);
373 if (dev->flags & IFF_UP)
379 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
381 const struct Qdisc_class_ops *cops;
387 while ((parentid = sch->parent)) {
388 sch = qdisc_lookup(sch->dev, TC_H_MAJ(parentid));
390 WARN_ON(parentid != TC_H_ROOT);
393 cops = sch->ops->cl_ops;
394 if (cops->qlen_notify) {
395 cl = cops->get(sch, parentid);
396 cops->qlen_notify(sch, cl);
402 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
404 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
407 Old qdisc is not destroyed but returned in *old.
410 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
412 struct Qdisc *new, struct Qdisc **old)
415 struct Qdisc *q = *old;
418 if (parent == NULL) {
419 if (q && q->flags&TCQ_F_INGRESS) {
420 *old = dev_graft_qdisc(dev, q);
422 *old = dev_graft_qdisc(dev, new);
425 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
430 unsigned long cl = cops->get(parent, classid);
432 err = cops->graft(parent, cl, new, old);
433 cops->put(parent, cl);
441 Allocate and initialize new qdisc.
443 Parameters are passed via opt.
446 static struct Qdisc *
447 qdisc_create(struct net_device *dev, u32 parent, u32 handle,
448 struct nlattr **tca, int *errp)
451 struct nlattr *kind = tca[TCA_KIND];
453 struct Qdisc_ops *ops;
455 ops = qdisc_lookup_ops(kind);
457 if (ops == NULL && kind != NULL) {
459 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
460 /* We dropped the RTNL semaphore in order to
461 * perform the module load. So, even if we
462 * succeeded in loading the module we have to
463 * tell the caller to replay the request. We
464 * indicate this using -EAGAIN.
465 * We replay the request because the device may
466 * go away in the mean time.
469 request_module("sch_%s", name);
471 ops = qdisc_lookup_ops(kind);
473 /* We will try again qdisc_lookup_ops,
474 * so don't keep a reference.
476 module_put(ops->owner);
488 sch = qdisc_alloc(dev, ops);
494 sch->parent = parent;
496 if (handle == TC_H_INGRESS) {
497 sch->flags |= TCQ_F_INGRESS;
498 sch->stats_lock = &dev->ingress_lock;
499 handle = TC_H_MAKE(TC_H_INGRESS, 0);
501 sch->stats_lock = &dev->queue_lock;
503 handle = qdisc_alloc_handle(dev);
510 sch->handle = handle;
512 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
514 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
519 * Any broken qdiscs that would require
520 * a ops->reset() here? The qdisc was never
521 * in action so it shouldn't be necessary.
528 qdisc_lock_tree(dev);
529 list_add_tail(&sch->list, &dev->qdisc_list);
530 qdisc_unlock_tree(dev);
536 kfree((char *) sch - sch->padded);
538 module_put(ops->owner);
544 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
546 if (tca[TCA_OPTIONS]) {
549 if (sch->ops->change == NULL)
551 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
556 gen_replace_estimator(&sch->bstats, &sch->rate_est,
557 sch->stats_lock, tca[TCA_RATE]);
561 struct check_loop_arg
563 struct qdisc_walker w;
568 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
570 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
572 struct check_loop_arg arg;
574 if (q->ops->cl_ops == NULL)
577 arg.w.stop = arg.w.skip = arg.w.count = 0;
578 arg.w.fn = check_loop_fn;
581 q->ops->cl_ops->walk(q, &arg.w);
582 return arg.w.stop ? -ELOOP : 0;
586 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
589 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
590 struct check_loop_arg *arg = (struct check_loop_arg *)w;
592 leaf = cops->leaf(q, cl);
594 if (leaf == arg->p || arg->depth > 7)
596 return check_loop(leaf, arg->p, arg->depth + 1);
605 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
607 struct net *net = skb->sk->sk_net;
608 struct tcmsg *tcm = NLMSG_DATA(n);
609 struct nlattr *tca[TCA_MAX + 1];
610 struct net_device *dev;
611 u32 clid = tcm->tcm_parent;
612 struct Qdisc *q = NULL;
613 struct Qdisc *p = NULL;
616 if (net != &init_net)
619 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
622 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
627 if (clid != TC_H_ROOT) {
628 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
629 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
631 q = qdisc_leaf(p, clid);
632 } else { /* ingress */
633 q = dev->qdisc_ingress;
636 q = dev->qdisc_sleeping;
641 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
644 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
648 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
651 if (n->nlmsg_type == RTM_DELQDISC) {
656 if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0)
659 qdisc_notify(skb, n, clid, q, NULL);
660 qdisc_lock_tree(dev);
662 qdisc_unlock_tree(dev);
665 qdisc_notify(skb, n, clid, NULL, q);
674 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
676 struct net *net = skb->sk->sk_net;
678 struct nlattr *tca[TCA_MAX + 1];
679 struct net_device *dev;
684 if (net != &init_net)
688 /* Reinit, just in case something touches this. */
690 clid = tcm->tcm_parent;
693 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
696 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
701 if (clid != TC_H_ROOT) {
702 if (clid != TC_H_INGRESS) {
703 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
705 q = qdisc_leaf(p, clid);
706 } else { /*ingress */
707 q = dev->qdisc_ingress;
710 q = dev->qdisc_sleeping;
713 /* It may be default qdisc, ignore it */
714 if (q && q->handle == 0)
717 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
718 if (tcm->tcm_handle) {
719 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
721 if (TC_H_MIN(tcm->tcm_handle))
723 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
725 if (n->nlmsg_flags&NLM_F_EXCL)
727 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
730 (p && check_loop(q, p, 0)))
732 atomic_inc(&q->refcnt);
738 /* This magic test requires explanation.
740 * We know, that some child q is already
741 * attached to this parent and have choice:
742 * either to change it or to create/graft new one.
744 * 1. We are allowed to create/graft only
745 * if CREATE and REPLACE flags are set.
747 * 2. If EXCL is set, requestor wanted to say,
748 * that qdisc tcm_handle is not expected
749 * to exist, so that we choose create/graft too.
751 * 3. The last case is when no flags are set.
752 * Alas, it is sort of hole in API, we
753 * cannot decide what to do unambiguously.
754 * For now we select create/graft, if
755 * user gave KIND, which does not match existing.
757 if ((n->nlmsg_flags&NLM_F_CREATE) &&
758 (n->nlmsg_flags&NLM_F_REPLACE) &&
759 ((n->nlmsg_flags&NLM_F_EXCL) ||
761 nla_strcmp(tca[TCA_KIND], q->ops->id))))
766 if (!tcm->tcm_handle)
768 q = qdisc_lookup(dev, tcm->tcm_handle);
771 /* Change qdisc parameters */
774 if (n->nlmsg_flags&NLM_F_EXCL)
776 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
778 err = qdisc_change(q, tca);
780 qdisc_notify(skb, n, clid, NULL, q);
784 if (!(n->nlmsg_flags&NLM_F_CREATE))
786 if (clid == TC_H_INGRESS)
787 q = qdisc_create(dev, tcm->tcm_parent, tcm->tcm_parent,
790 q = qdisc_create(dev, tcm->tcm_parent, tcm->tcm_handle,
800 struct Qdisc *old_q = NULL;
801 err = qdisc_graft(dev, p, clid, q, &old_q);
804 qdisc_lock_tree(dev);
806 qdisc_unlock_tree(dev);
810 qdisc_notify(skb, n, clid, old_q, q);
812 qdisc_lock_tree(dev);
813 qdisc_destroy(old_q);
814 qdisc_unlock_tree(dev);
820 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
821 u32 pid, u32 seq, u16 flags, int event)
824 struct nlmsghdr *nlh;
825 unsigned char *b = skb_tail_pointer(skb);
828 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
829 tcm = NLMSG_DATA(nlh);
830 tcm->tcm_family = AF_UNSPEC;
833 tcm->tcm_ifindex = q->dev->ifindex;
834 tcm->tcm_parent = clid;
835 tcm->tcm_handle = q->handle;
836 tcm->tcm_info = atomic_read(&q->refcnt);
837 NLA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id);
838 if (q->ops->dump && q->ops->dump(q, skb) < 0)
839 goto nla_put_failure;
840 q->qstats.qlen = q->q.qlen;
842 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
843 TCA_XSTATS, q->stats_lock, &d) < 0)
844 goto nla_put_failure;
846 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
847 goto nla_put_failure;
849 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
850 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
851 gnet_stats_copy_queue(&d, &q->qstats) < 0)
852 goto nla_put_failure;
854 if (gnet_stats_finish_copy(&d) < 0)
855 goto nla_put_failure;
857 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
866 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
867 u32 clid, struct Qdisc *old, struct Qdisc *new)
870 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
872 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
876 if (old && old->handle) {
877 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
881 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
886 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
893 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
895 struct net *net = skb->sk->sk_net;
898 struct net_device *dev;
901 if (net != &init_net)
905 s_q_idx = q_idx = cb->args[1];
906 read_lock(&dev_base_lock);
908 for_each_netdev(&init_net, dev) {
914 list_for_each_entry(q, &dev->qdisc_list, list) {
915 if (q_idx < s_q_idx) {
919 if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
920 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
929 read_unlock(&dev_base_lock);
939 /************************************************
940 * Traffic classes manipulation. *
941 ************************************************/
945 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
947 struct net *net = skb->sk->sk_net;
948 struct tcmsg *tcm = NLMSG_DATA(n);
949 struct nlattr *tca[TCA_MAX + 1];
950 struct net_device *dev;
951 struct Qdisc *q = NULL;
952 const struct Qdisc_class_ops *cops;
953 unsigned long cl = 0;
954 unsigned long new_cl;
955 u32 pid = tcm->tcm_parent;
956 u32 clid = tcm->tcm_handle;
957 u32 qid = TC_H_MAJ(clid);
960 if (net != &init_net)
963 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
966 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
971 parent == TC_H_UNSPEC - unspecified parent.
972 parent == TC_H_ROOT - class is root, which has no parent.
973 parent == X:0 - parent is root class.
974 parent == X:Y - parent is a node in hierarchy.
975 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
977 handle == 0:0 - generate handle from kernel pool.
978 handle == 0:Y - class is X:Y, where X:0 is qdisc.
979 handle == X:Y - clear.
980 handle == X:0 - root class.
983 /* Step 1. Determine qdisc handle X:0 */
985 if (pid != TC_H_ROOT) {
986 u32 qid1 = TC_H_MAJ(pid);
989 /* If both majors are known, they must be identical. */
995 qid = dev->qdisc_sleeping->handle;
997 /* Now qid is genuine qdisc handle consistent
998 both with parent and child.
1000 TC_H_MAJ(pid) still may be unspecified, complete it now.
1003 pid = TC_H_MAKE(qid, pid);
1006 qid = dev->qdisc_sleeping->handle;
1009 /* OK. Locate qdisc */
1010 if ((q = qdisc_lookup(dev, qid)) == NULL)
1013 /* An check that it supports classes */
1014 cops = q->ops->cl_ops;
1018 /* Now try to get class */
1020 if (pid == TC_H_ROOT)
1023 clid = TC_H_MAKE(qid, clid);
1026 cl = cops->get(q, clid);
1030 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1033 switch (n->nlmsg_type) {
1036 if (n->nlmsg_flags&NLM_F_EXCL)
1040 err = cops->delete(q, cl);
1042 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1045 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1054 err = cops->change(q, clid, pid, tca, &new_cl);
1056 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1066 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1068 u32 pid, u32 seq, u16 flags, int event)
1071 struct nlmsghdr *nlh;
1072 unsigned char *b = skb_tail_pointer(skb);
1074 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1076 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1077 tcm = NLMSG_DATA(nlh);
1078 tcm->tcm_family = AF_UNSPEC;
1079 tcm->tcm_ifindex = q->dev->ifindex;
1080 tcm->tcm_parent = q->handle;
1081 tcm->tcm_handle = q->handle;
1083 NLA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id);
1084 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1085 goto nla_put_failure;
1087 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1088 TCA_XSTATS, q->stats_lock, &d) < 0)
1089 goto nla_put_failure;
1091 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1092 goto nla_put_failure;
1094 if (gnet_stats_finish_copy(&d) < 0)
1095 goto nla_put_failure;
1097 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1106 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1107 struct Qdisc *q, unsigned long cl, int event)
1109 struct sk_buff *skb;
1110 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1112 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1116 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1121 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1124 struct qdisc_dump_args
1126 struct qdisc_walker w;
1127 struct sk_buff *skb;
1128 struct netlink_callback *cb;
1131 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1133 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1135 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1136 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1139 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1141 struct net *net = skb->sk->sk_net;
1144 struct net_device *dev;
1146 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1147 struct qdisc_dump_args arg;
1149 if (net != &init_net)
1152 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1154 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1160 list_for_each_entry(q, &dev->qdisc_list, list) {
1161 if (t < s_t || !q->ops->cl_ops ||
1163 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1168 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1169 arg.w.fn = qdisc_class_dump;
1173 arg.w.skip = cb->args[1];
1175 q->ops->cl_ops->walk(q, &arg.w);
1176 cb->args[1] = arg.w.count;
1188 /* Main classifier routine: scans classifier chain attached
1189 to this qdisc, (optionally) tests for protocol and asks
1190 specific classifiers.
1192 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1193 struct tcf_result *res)
1195 __be16 protocol = skb->protocol;
1198 for (; tp; tp = tp->next) {
1199 if ((tp->protocol == protocol ||
1200 tp->protocol == htons(ETH_P_ALL)) &&
1201 (err = tp->classify(skb, tp, res)) >= 0) {
1202 #ifdef CONFIG_NET_CLS_ACT
1203 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1204 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1211 EXPORT_SYMBOL(tc_classify_compat);
1213 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1214 struct tcf_result *res)
1218 #ifdef CONFIG_NET_CLS_ACT
1219 struct tcf_proto *otp = tp;
1222 protocol = skb->protocol;
1224 err = tc_classify_compat(skb, tp, res);
1225 #ifdef CONFIG_NET_CLS_ACT
1226 if (err == TC_ACT_RECLASSIFY) {
1227 u32 verd = G_TC_VERD(skb->tc_verd);
1230 if (verd++ >= MAX_REC_LOOP) {
1231 printk("rule prio %u protocol %02x reclassify loop, "
1233 tp->prio&0xffff, ntohs(tp->protocol));
1236 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1242 EXPORT_SYMBOL(tc_classify);
1244 void tcf_destroy(struct tcf_proto *tp)
1246 tp->ops->destroy(tp);
1247 module_put(tp->ops->owner);
1251 void tcf_destroy_chain(struct tcf_proto *fl)
1253 struct tcf_proto *tp;
1255 while ((tp = fl) != NULL) {
1260 EXPORT_SYMBOL(tcf_destroy_chain);
1262 #ifdef CONFIG_PROC_FS
1263 static int psched_show(struct seq_file *seq, void *v)
1267 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1268 seq_printf(seq, "%08x %08x %08x %08x\n",
1269 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1271 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1276 static int psched_open(struct inode *inode, struct file *file)
1278 return single_open(file, psched_show, PDE(inode)->data);
1281 static const struct file_operations psched_fops = {
1282 .owner = THIS_MODULE,
1283 .open = psched_open,
1285 .llseek = seq_lseek,
1286 .release = single_release,
1290 static int __init pktsched_init(void)
1292 register_qdisc(&pfifo_qdisc_ops);
1293 register_qdisc(&bfifo_qdisc_ops);
1294 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1296 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1297 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1298 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1299 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1300 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1301 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1306 subsys_initcall(pktsched_init);