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.
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,
284 struct net_device *dev = qdisc_dev(wd->qdisc);
286 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
290 return HRTIMER_NORESTART;
293 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
295 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
296 wd->timer.function = qdisc_watchdog;
299 EXPORT_SYMBOL(qdisc_watchdog_init);
301 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
305 wd->qdisc->flags |= TCQ_F_THROTTLED;
306 time = ktime_set(0, 0);
307 time = ktime_add_ns(time, PSCHED_US2NS(expires));
308 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
310 EXPORT_SYMBOL(qdisc_watchdog_schedule);
312 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
314 hrtimer_cancel(&wd->timer);
315 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
317 EXPORT_SYMBOL(qdisc_watchdog_cancel);
319 struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
321 unsigned int size = n * sizeof(struct hlist_head), i;
322 struct hlist_head *h;
324 if (size <= PAGE_SIZE)
325 h = kmalloc(size, GFP_KERNEL);
327 h = (struct hlist_head *)
328 __get_free_pages(GFP_KERNEL, get_order(size));
331 for (i = 0; i < n; i++)
332 INIT_HLIST_HEAD(&h[i]);
337 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
339 unsigned int size = n * sizeof(struct hlist_head);
341 if (size <= PAGE_SIZE)
344 free_pages((unsigned long)h, get_order(size));
347 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
349 struct Qdisc_class_common *cl;
350 struct hlist_node *n, *next;
351 struct hlist_head *nhash, *ohash;
352 unsigned int nsize, nmask, osize;
355 /* Rehash when load factor exceeds 0.75 */
356 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
358 nsize = clhash->hashsize * 2;
360 nhash = qdisc_class_hash_alloc(nsize);
364 ohash = clhash->hash;
365 osize = clhash->hashsize;
368 for (i = 0; i < osize; i++) {
369 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
370 h = qdisc_class_hash(cl->classid, nmask);
371 hlist_add_head(&cl->hnode, &nhash[h]);
374 clhash->hash = nhash;
375 clhash->hashsize = nsize;
376 clhash->hashmask = nmask;
377 sch_tree_unlock(sch);
379 qdisc_class_hash_free(ohash, osize);
381 EXPORT_SYMBOL(qdisc_class_hash_grow);
383 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
385 unsigned int size = 4;
387 clhash->hash = qdisc_class_hash_alloc(size);
388 if (clhash->hash == NULL)
390 clhash->hashsize = size;
391 clhash->hashmask = size - 1;
392 clhash->hashelems = 0;
395 EXPORT_SYMBOL(qdisc_class_hash_init);
397 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
399 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
401 EXPORT_SYMBOL(qdisc_class_hash_destroy);
403 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
404 struct Qdisc_class_common *cl)
408 INIT_HLIST_NODE(&cl->hnode);
409 h = qdisc_class_hash(cl->classid, clhash->hashmask);
410 hlist_add_head(&cl->hnode, &clhash->hash[h]);
413 EXPORT_SYMBOL(qdisc_class_hash_insert);
415 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
416 struct Qdisc_class_common *cl)
418 hlist_del(&cl->hnode);
421 EXPORT_SYMBOL(qdisc_class_hash_remove);
423 /* Allocate an unique handle from space managed by kernel */
425 static u32 qdisc_alloc_handle(struct net_device *dev)
428 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
431 autohandle += TC_H_MAKE(0x10000U, 0);
432 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
433 autohandle = TC_H_MAKE(0x80000000U, 0);
434 } while (qdisc_lookup(dev, autohandle) && --i > 0);
436 return i>0 ? autohandle : 0;
439 /* Attach toplevel qdisc to device dev */
441 static struct Qdisc *
442 dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc)
444 struct Qdisc *oqdisc;
446 if (dev->flags & IFF_UP)
449 qdisc_lock_tree(dev);
450 if (qdisc && qdisc->flags&TCQ_F_INGRESS) {
451 oqdisc = dev->qdisc_ingress;
452 /* Prune old scheduler */
453 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) {
456 dev->qdisc_ingress = NULL;
458 dev->qdisc_ingress = qdisc;
463 oqdisc = dev->qdisc_sleeping;
465 /* Prune old scheduler */
466 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
469 /* ... and graft new one */
472 dev->qdisc_sleeping = qdisc;
473 dev->qdisc = &noop_qdisc;
476 qdisc_unlock_tree(dev);
478 if (dev->flags & IFF_UP)
484 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
486 const struct Qdisc_class_ops *cops;
492 while ((parentid = sch->parent)) {
493 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
496 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
498 WARN_ON(parentid != TC_H_ROOT);
501 cops = sch->ops->cl_ops;
502 if (cops->qlen_notify) {
503 cl = cops->get(sch, parentid);
504 cops->qlen_notify(sch, cl);
510 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
512 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
515 Old qdisc is not destroyed but returned in *old.
518 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
520 struct Qdisc *new, struct Qdisc **old)
523 struct Qdisc *q = *old;
526 if (parent == NULL) {
527 if (q && q->flags&TCQ_F_INGRESS) {
528 *old = dev_graft_qdisc(dev, q);
530 *old = dev_graft_qdisc(dev, new);
533 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
538 unsigned long cl = cops->get(parent, classid);
540 err = cops->graft(parent, cl, new, old);
541 cops->put(parent, cl);
549 Allocate and initialize new qdisc.
551 Parameters are passed via opt.
554 static struct Qdisc *
555 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
556 u32 parent, u32 handle, struct nlattr **tca, int *errp)
559 struct nlattr *kind = tca[TCA_KIND];
561 struct Qdisc_ops *ops;
563 ops = qdisc_lookup_ops(kind);
565 if (ops == NULL && kind != NULL) {
567 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
568 /* We dropped the RTNL semaphore in order to
569 * perform the module load. So, even if we
570 * succeeded in loading the module we have to
571 * tell the caller to replay the request. We
572 * indicate this using -EAGAIN.
573 * We replay the request because the device may
574 * go away in the mean time.
577 request_module("sch_%s", name);
579 ops = qdisc_lookup_ops(kind);
581 /* We will try again qdisc_lookup_ops,
582 * so don't keep a reference.
584 module_put(ops->owner);
596 sch = qdisc_alloc(dev_queue, ops);
602 sch->parent = parent;
604 if (handle == TC_H_INGRESS) {
605 sch->flags |= TCQ_F_INGRESS;
606 sch->stats_lock = &dev->ingress_lock;
607 handle = TC_H_MAKE(TC_H_INGRESS, 0);
609 sch->stats_lock = &dev->queue_lock;
611 handle = qdisc_alloc_handle(dev);
618 sch->handle = handle;
620 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
622 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
627 * Any broken qdiscs that would require
628 * a ops->reset() here? The qdisc was never
629 * in action so it shouldn't be necessary.
636 qdisc_lock_tree(dev);
637 list_add_tail(&sch->list, &dev->qdisc_list);
638 qdisc_unlock_tree(dev);
644 kfree((char *) sch - sch->padded);
646 module_put(ops->owner);
652 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
654 if (tca[TCA_OPTIONS]) {
657 if (sch->ops->change == NULL)
659 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
664 gen_replace_estimator(&sch->bstats, &sch->rate_est,
665 sch->stats_lock, tca[TCA_RATE]);
669 struct check_loop_arg
671 struct qdisc_walker w;
676 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
678 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
680 struct check_loop_arg arg;
682 if (q->ops->cl_ops == NULL)
685 arg.w.stop = arg.w.skip = arg.w.count = 0;
686 arg.w.fn = check_loop_fn;
689 q->ops->cl_ops->walk(q, &arg.w);
690 return arg.w.stop ? -ELOOP : 0;
694 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
697 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
698 struct check_loop_arg *arg = (struct check_loop_arg *)w;
700 leaf = cops->leaf(q, cl);
702 if (leaf == arg->p || arg->depth > 7)
704 return check_loop(leaf, arg->p, arg->depth + 1);
713 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
715 struct net *net = sock_net(skb->sk);
716 struct tcmsg *tcm = NLMSG_DATA(n);
717 struct nlattr *tca[TCA_MAX + 1];
718 struct net_device *dev;
719 u32 clid = tcm->tcm_parent;
720 struct Qdisc *q = NULL;
721 struct Qdisc *p = NULL;
724 if (net != &init_net)
727 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
730 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
735 if (clid != TC_H_ROOT) {
736 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
737 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
739 q = qdisc_leaf(p, clid);
740 } else { /* ingress */
741 q = dev->qdisc_ingress;
744 q = dev->qdisc_sleeping;
749 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
752 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
756 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
759 if (n->nlmsg_type == RTM_DELQDISC) {
764 if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0)
767 qdisc_notify(skb, n, clid, q, NULL);
768 qdisc_lock_tree(dev);
770 qdisc_unlock_tree(dev);
773 qdisc_notify(skb, n, clid, NULL, q);
782 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
784 struct net *net = sock_net(skb->sk);
786 struct nlattr *tca[TCA_MAX + 1];
787 struct net_device *dev;
792 if (net != &init_net)
796 /* Reinit, just in case something touches this. */
798 clid = tcm->tcm_parent;
801 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
804 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
809 if (clid != TC_H_ROOT) {
810 if (clid != TC_H_INGRESS) {
811 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
813 q = qdisc_leaf(p, clid);
814 } else { /*ingress */
815 q = dev->qdisc_ingress;
818 q = dev->qdisc_sleeping;
821 /* It may be default qdisc, ignore it */
822 if (q && q->handle == 0)
825 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
826 if (tcm->tcm_handle) {
827 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
829 if (TC_H_MIN(tcm->tcm_handle))
831 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
833 if (n->nlmsg_flags&NLM_F_EXCL)
835 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
838 (p && check_loop(q, p, 0)))
840 atomic_inc(&q->refcnt);
846 /* This magic test requires explanation.
848 * We know, that some child q is already
849 * attached to this parent and have choice:
850 * either to change it or to create/graft new one.
852 * 1. We are allowed to create/graft only
853 * if CREATE and REPLACE flags are set.
855 * 2. If EXCL is set, requestor wanted to say,
856 * that qdisc tcm_handle is not expected
857 * to exist, so that we choose create/graft too.
859 * 3. The last case is when no flags are set.
860 * Alas, it is sort of hole in API, we
861 * cannot decide what to do unambiguously.
862 * For now we select create/graft, if
863 * user gave KIND, which does not match existing.
865 if ((n->nlmsg_flags&NLM_F_CREATE) &&
866 (n->nlmsg_flags&NLM_F_REPLACE) &&
867 ((n->nlmsg_flags&NLM_F_EXCL) ||
869 nla_strcmp(tca[TCA_KIND], q->ops->id))))
874 if (!tcm->tcm_handle)
876 q = qdisc_lookup(dev, tcm->tcm_handle);
879 /* Change qdisc parameters */
882 if (n->nlmsg_flags&NLM_F_EXCL)
884 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
886 err = qdisc_change(q, tca);
888 qdisc_notify(skb, n, clid, NULL, q);
892 if (!(n->nlmsg_flags&NLM_F_CREATE))
894 if (clid == TC_H_INGRESS)
895 q = qdisc_create(dev, &dev->rx_queue,
896 tcm->tcm_parent, tcm->tcm_parent,
899 q = qdisc_create(dev, &dev->tx_queue,
900 tcm->tcm_parent, tcm->tcm_handle,
910 struct Qdisc *old_q = NULL;
911 err = qdisc_graft(dev, p, clid, q, &old_q);
914 qdisc_lock_tree(dev);
916 qdisc_unlock_tree(dev);
920 qdisc_notify(skb, n, clid, old_q, q);
922 qdisc_lock_tree(dev);
923 qdisc_destroy(old_q);
924 qdisc_unlock_tree(dev);
930 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
931 u32 pid, u32 seq, u16 flags, int event)
934 struct nlmsghdr *nlh;
935 unsigned char *b = skb_tail_pointer(skb);
938 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
939 tcm = NLMSG_DATA(nlh);
940 tcm->tcm_family = AF_UNSPEC;
943 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
944 tcm->tcm_parent = clid;
945 tcm->tcm_handle = q->handle;
946 tcm->tcm_info = atomic_read(&q->refcnt);
947 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
948 if (q->ops->dump && q->ops->dump(q, skb) < 0)
949 goto nla_put_failure;
950 q->qstats.qlen = q->q.qlen;
952 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
953 TCA_XSTATS, q->stats_lock, &d) < 0)
954 goto nla_put_failure;
956 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
957 goto nla_put_failure;
959 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
960 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
961 gnet_stats_copy_queue(&d, &q->qstats) < 0)
962 goto nla_put_failure;
964 if (gnet_stats_finish_copy(&d) < 0)
965 goto nla_put_failure;
967 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
976 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
977 u32 clid, struct Qdisc *old, struct Qdisc *new)
980 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
982 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
986 if (old && old->handle) {
987 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
991 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
996 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1003 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1005 struct net *net = sock_net(skb->sk);
1008 struct net_device *dev;
1011 if (net != &init_net)
1014 s_idx = cb->args[0];
1015 s_q_idx = q_idx = cb->args[1];
1016 read_lock(&dev_base_lock);
1018 for_each_netdev(&init_net, dev) {
1024 list_for_each_entry(q, &dev->qdisc_list, list) {
1025 if (q_idx < s_q_idx) {
1029 if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1030 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1039 read_unlock(&dev_base_lock);
1042 cb->args[1] = q_idx;
1049 /************************************************
1050 * Traffic classes manipulation. *
1051 ************************************************/
1055 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1057 struct net *net = sock_net(skb->sk);
1058 struct tcmsg *tcm = NLMSG_DATA(n);
1059 struct nlattr *tca[TCA_MAX + 1];
1060 struct net_device *dev;
1061 struct Qdisc *q = NULL;
1062 const struct Qdisc_class_ops *cops;
1063 unsigned long cl = 0;
1064 unsigned long new_cl;
1065 u32 pid = tcm->tcm_parent;
1066 u32 clid = tcm->tcm_handle;
1067 u32 qid = TC_H_MAJ(clid);
1070 if (net != &init_net)
1073 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1076 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1081 parent == TC_H_UNSPEC - unspecified parent.
1082 parent == TC_H_ROOT - class is root, which has no parent.
1083 parent == X:0 - parent is root class.
1084 parent == X:Y - parent is a node in hierarchy.
1085 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1087 handle == 0:0 - generate handle from kernel pool.
1088 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1089 handle == X:Y - clear.
1090 handle == X:0 - root class.
1093 /* Step 1. Determine qdisc handle X:0 */
1095 if (pid != TC_H_ROOT) {
1096 u32 qid1 = TC_H_MAJ(pid);
1099 /* If both majors are known, they must be identical. */
1104 } else if (qid == 0)
1105 qid = dev->qdisc_sleeping->handle;
1107 /* Now qid is genuine qdisc handle consistent
1108 both with parent and child.
1110 TC_H_MAJ(pid) still may be unspecified, complete it now.
1113 pid = TC_H_MAKE(qid, pid);
1116 qid = dev->qdisc_sleeping->handle;
1119 /* OK. Locate qdisc */
1120 if ((q = qdisc_lookup(dev, qid)) == NULL)
1123 /* An check that it supports classes */
1124 cops = q->ops->cl_ops;
1128 /* Now try to get class */
1130 if (pid == TC_H_ROOT)
1133 clid = TC_H_MAKE(qid, clid);
1136 cl = cops->get(q, clid);
1140 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1143 switch (n->nlmsg_type) {
1146 if (n->nlmsg_flags&NLM_F_EXCL)
1150 err = cops->delete(q, cl);
1152 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1155 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1164 err = cops->change(q, clid, pid, tca, &new_cl);
1166 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1176 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1178 u32 pid, u32 seq, u16 flags, int event)
1181 struct nlmsghdr *nlh;
1182 unsigned char *b = skb_tail_pointer(skb);
1184 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1186 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1187 tcm = NLMSG_DATA(nlh);
1188 tcm->tcm_family = AF_UNSPEC;
1189 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1190 tcm->tcm_parent = q->handle;
1191 tcm->tcm_handle = q->handle;
1193 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1194 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1195 goto nla_put_failure;
1197 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1198 TCA_XSTATS, q->stats_lock, &d) < 0)
1199 goto nla_put_failure;
1201 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1202 goto nla_put_failure;
1204 if (gnet_stats_finish_copy(&d) < 0)
1205 goto nla_put_failure;
1207 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1216 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1217 struct Qdisc *q, unsigned long cl, int event)
1219 struct sk_buff *skb;
1220 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1222 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1226 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1231 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1234 struct qdisc_dump_args
1236 struct qdisc_walker w;
1237 struct sk_buff *skb;
1238 struct netlink_callback *cb;
1241 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1243 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1245 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1246 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1249 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1251 struct net *net = sock_net(skb->sk);
1254 struct net_device *dev;
1256 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1257 struct qdisc_dump_args arg;
1259 if (net != &init_net)
1262 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1264 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1270 list_for_each_entry(q, &dev->qdisc_list, list) {
1271 if (t < s_t || !q->ops->cl_ops ||
1273 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1278 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1279 arg.w.fn = qdisc_class_dump;
1283 arg.w.skip = cb->args[1];
1285 q->ops->cl_ops->walk(q, &arg.w);
1286 cb->args[1] = arg.w.count;
1298 /* Main classifier routine: scans classifier chain attached
1299 to this qdisc, (optionally) tests for protocol and asks
1300 specific classifiers.
1302 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1303 struct tcf_result *res)
1305 __be16 protocol = skb->protocol;
1308 for (; tp; tp = tp->next) {
1309 if ((tp->protocol == protocol ||
1310 tp->protocol == htons(ETH_P_ALL)) &&
1311 (err = tp->classify(skb, tp, res)) >= 0) {
1312 #ifdef CONFIG_NET_CLS_ACT
1313 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1314 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1321 EXPORT_SYMBOL(tc_classify_compat);
1323 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1324 struct tcf_result *res)
1328 #ifdef CONFIG_NET_CLS_ACT
1329 struct tcf_proto *otp = tp;
1332 protocol = skb->protocol;
1334 err = tc_classify_compat(skb, tp, res);
1335 #ifdef CONFIG_NET_CLS_ACT
1336 if (err == TC_ACT_RECLASSIFY) {
1337 u32 verd = G_TC_VERD(skb->tc_verd);
1340 if (verd++ >= MAX_REC_LOOP) {
1341 printk("rule prio %u protocol %02x reclassify loop, "
1343 tp->prio&0xffff, ntohs(tp->protocol));
1346 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1352 EXPORT_SYMBOL(tc_classify);
1354 void tcf_destroy(struct tcf_proto *tp)
1356 tp->ops->destroy(tp);
1357 module_put(tp->ops->owner);
1361 void tcf_destroy_chain(struct tcf_proto **fl)
1363 struct tcf_proto *tp;
1365 while ((tp = *fl) != NULL) {
1370 EXPORT_SYMBOL(tcf_destroy_chain);
1372 #ifdef CONFIG_PROC_FS
1373 static int psched_show(struct seq_file *seq, void *v)
1377 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1378 seq_printf(seq, "%08x %08x %08x %08x\n",
1379 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1381 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1386 static int psched_open(struct inode *inode, struct file *file)
1388 return single_open(file, psched_show, PDE(inode)->data);
1391 static const struct file_operations psched_fops = {
1392 .owner = THIS_MODULE,
1393 .open = psched_open,
1395 .llseek = seq_lseek,
1396 .release = single_release,
1400 static int __init pktsched_init(void)
1402 register_qdisc(&pfifo_qdisc_ops);
1403 register_qdisc(&bfifo_qdisc_ops);
1404 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1406 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1407 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1408 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1409 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1410 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1411 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1416 subsys_initcall(pktsched_init);