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>
30 #include <linux/lockdep.h>
32 #include <net/net_namespace.h>
34 #include <net/netlink.h>
35 #include <net/pkt_sched.h>
37 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid,
38 struct Qdisc *old, struct Qdisc *new);
39 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
40 struct Qdisc *q, unsigned long cl, int event);
47 This file consists of two interrelated parts:
49 1. queueing disciplines manager frontend.
50 2. traffic classes manager frontend.
52 Generally, queueing discipline ("qdisc") is a black box,
53 which is able to enqueue packets and to dequeue them (when
54 device is ready to send something) in order and at times
55 determined by algorithm hidden in it.
57 qdisc's are divided to two categories:
58 - "queues", which have no internal structure visible from outside.
59 - "schedulers", which split all the packets to "traffic classes",
60 using "packet classifiers" (look at cls_api.c)
62 In turn, classes may have child qdiscs (as rule, queues)
63 attached to them etc. etc. etc.
65 The goal of the routines in this file is to translate
66 information supplied by user in the form of handles
67 to more intelligible for kernel form, to make some sanity
68 checks and part of work, which is common to all qdiscs
69 and to provide rtnetlink notifications.
71 All real intelligent work is done inside qdisc modules.
75 Every discipline has two major routines: enqueue and dequeue.
79 dequeue usually returns a skb to send. It is allowed to return NULL,
80 but it does not mean that queue is empty, it just means that
81 discipline does not want to send anything this time.
82 Queue is really empty if q->q.qlen == 0.
83 For complicated disciplines with multiple queues q->q is not
84 real packet queue, but however q->q.qlen must be valid.
88 enqueue returns 0, if packet was enqueued successfully.
89 If packet (this one or another one) was dropped, it returns
91 NET_XMIT_DROP - this packet dropped
92 Expected action: do not backoff, but wait until queue will clear.
93 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
94 Expected action: backoff or ignore
95 NET_XMIT_POLICED - dropped by police.
96 Expected action: backoff or error to real-time apps.
102 requeues once dequeued packet. It is used for non-standard or
103 just buggy devices, which can defer output even if netif_queue_stopped()=0.
107 returns qdisc to initial state: purge all buffers, clear all
108 timers, counters (except for statistics) etc.
112 initializes newly created qdisc.
116 destroys resources allocated by init and during lifetime of qdisc.
120 changes qdisc parameters.
123 /* Protects list of registered TC modules. It is pure SMP lock. */
124 static DEFINE_RWLOCK(qdisc_mod_lock);
127 /************************************************
128 * Queueing disciplines manipulation. *
129 ************************************************/
132 /* The list of all installed queueing disciplines. */
134 static struct Qdisc_ops *qdisc_base;
136 /* Register/uregister queueing discipline */
138 int register_qdisc(struct Qdisc_ops *qops)
140 struct Qdisc_ops *q, **qp;
143 write_lock(&qdisc_mod_lock);
144 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
145 if (!strcmp(qops->id, q->id))
148 if (qops->enqueue == NULL)
149 qops->enqueue = noop_qdisc_ops.enqueue;
150 if (qops->requeue == NULL)
151 qops->requeue = noop_qdisc_ops.requeue;
152 if (qops->dequeue == NULL)
153 qops->dequeue = noop_qdisc_ops.dequeue;
159 write_unlock(&qdisc_mod_lock);
162 EXPORT_SYMBOL(register_qdisc);
164 int unregister_qdisc(struct Qdisc_ops *qops)
166 struct Qdisc_ops *q, **qp;
169 write_lock(&qdisc_mod_lock);
170 for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next)
178 write_unlock(&qdisc_mod_lock);
181 EXPORT_SYMBOL(unregister_qdisc);
183 /* We know handle. Find qdisc among all qdisc's attached to device
184 (root qdisc, all its children, children of children etc.)
187 struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
191 if (!(root->flags & TCQ_F_BUILTIN) &&
192 root->handle == handle)
195 list_for_each_entry(q, &root->list, list) {
196 if (q->handle == handle)
202 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
206 for (i = 0; i < dev->num_tx_queues; i++) {
207 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
208 struct Qdisc *q, *txq_root = txq->qdisc_sleeping;
210 q = qdisc_match_from_root(txq_root, handle);
214 return qdisc_match_from_root(dev->rx_queue.qdisc_sleeping, handle);
217 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
221 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
225 cl = cops->get(p, classid);
229 leaf = cops->leaf(p, cl);
234 /* Find queueing discipline by name */
236 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
238 struct Qdisc_ops *q = NULL;
241 read_lock(&qdisc_mod_lock);
242 for (q = qdisc_base; q; q = q->next) {
243 if (nla_strcmp(kind, q->id) == 0) {
244 if (!try_module_get(q->owner))
249 read_unlock(&qdisc_mod_lock);
254 static struct qdisc_rate_table *qdisc_rtab_list;
256 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
258 struct qdisc_rate_table *rtab;
260 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
261 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
267 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
268 nla_len(tab) != TC_RTAB_SIZE)
271 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
275 memcpy(rtab->data, nla_data(tab), 1024);
276 rtab->next = qdisc_rtab_list;
277 qdisc_rtab_list = rtab;
281 EXPORT_SYMBOL(qdisc_get_rtab);
283 void qdisc_put_rtab(struct qdisc_rate_table *tab)
285 struct qdisc_rate_table *rtab, **rtabp;
287 if (!tab || --tab->refcnt)
290 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
298 EXPORT_SYMBOL(qdisc_put_rtab);
300 static LIST_HEAD(qdisc_stab_list);
301 static DEFINE_SPINLOCK(qdisc_stab_lock);
303 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
304 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
305 [TCA_STAB_DATA] = { .type = NLA_BINARY },
308 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
310 struct nlattr *tb[TCA_STAB_MAX + 1];
311 struct qdisc_size_table *stab;
312 struct tc_sizespec *s;
313 unsigned int tsize = 0;
317 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
320 if (!tb[TCA_STAB_BASE])
321 return ERR_PTR(-EINVAL);
323 s = nla_data(tb[TCA_STAB_BASE]);
326 if (!tb[TCA_STAB_DATA])
327 return ERR_PTR(-EINVAL);
328 tab = nla_data(tb[TCA_STAB_DATA]);
329 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
332 if (!s || tsize != s->tsize || (!tab && tsize > 0))
333 return ERR_PTR(-EINVAL);
335 spin_lock_bh(&qdisc_stab_lock);
337 list_for_each_entry(stab, &qdisc_stab_list, list) {
338 if (memcmp(&stab->szopts, s, sizeof(*s)))
340 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
343 spin_unlock_bh(&qdisc_stab_lock);
347 spin_unlock_bh(&qdisc_stab_lock);
349 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
351 return ERR_PTR(-ENOMEM);
356 memcpy(stab->data, tab, tsize * sizeof(u16));
358 spin_lock_bh(&qdisc_stab_lock);
359 list_add_tail(&stab->list, &qdisc_stab_list);
360 spin_unlock_bh(&qdisc_stab_lock);
365 void qdisc_put_stab(struct qdisc_size_table *tab)
370 spin_lock_bh(&qdisc_stab_lock);
372 if (--tab->refcnt == 0) {
373 list_del(&tab->list);
377 spin_unlock_bh(&qdisc_stab_lock);
379 EXPORT_SYMBOL(qdisc_put_stab);
381 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
385 nest = nla_nest_start(skb, TCA_STAB);
386 NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts);
387 nla_nest_end(skb, nest);
395 void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab)
399 pkt_len = skb->len + stab->szopts.overhead;
400 if (unlikely(!stab->szopts.tsize))
403 slot = pkt_len + stab->szopts.cell_align;
404 if (unlikely(slot < 0))
407 slot >>= stab->szopts.cell_log;
408 if (likely(slot < stab->szopts.tsize))
409 pkt_len = stab->data[slot];
411 pkt_len = stab->data[stab->szopts.tsize - 1] *
412 (slot / stab->szopts.tsize) +
413 stab->data[slot % stab->szopts.tsize];
415 pkt_len <<= stab->szopts.size_log;
417 if (unlikely(pkt_len < 1))
419 qdisc_skb_cb(skb)->pkt_len = pkt_len;
421 EXPORT_SYMBOL(qdisc_calculate_pkt_len);
423 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
425 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
428 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
430 __netif_schedule(qdisc_root(wd->qdisc));
432 return HRTIMER_NORESTART;
435 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
437 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
438 wd->timer.function = qdisc_watchdog;
441 EXPORT_SYMBOL(qdisc_watchdog_init);
443 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
447 wd->qdisc->flags |= TCQ_F_THROTTLED;
448 time = ktime_set(0, 0);
449 time = ktime_add_ns(time, PSCHED_US2NS(expires));
450 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
452 EXPORT_SYMBOL(qdisc_watchdog_schedule);
454 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
456 hrtimer_cancel(&wd->timer);
457 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
459 EXPORT_SYMBOL(qdisc_watchdog_cancel);
461 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
463 unsigned int size = n * sizeof(struct hlist_head), i;
464 struct hlist_head *h;
466 if (size <= PAGE_SIZE)
467 h = kmalloc(size, GFP_KERNEL);
469 h = (struct hlist_head *)
470 __get_free_pages(GFP_KERNEL, get_order(size));
473 for (i = 0; i < n; i++)
474 INIT_HLIST_HEAD(&h[i]);
479 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
481 unsigned int size = n * sizeof(struct hlist_head);
483 if (size <= PAGE_SIZE)
486 free_pages((unsigned long)h, get_order(size));
489 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
491 struct Qdisc_class_common *cl;
492 struct hlist_node *n, *next;
493 struct hlist_head *nhash, *ohash;
494 unsigned int nsize, nmask, osize;
497 /* Rehash when load factor exceeds 0.75 */
498 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
500 nsize = clhash->hashsize * 2;
502 nhash = qdisc_class_hash_alloc(nsize);
506 ohash = clhash->hash;
507 osize = clhash->hashsize;
510 for (i = 0; i < osize; i++) {
511 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
512 h = qdisc_class_hash(cl->classid, nmask);
513 hlist_add_head(&cl->hnode, &nhash[h]);
516 clhash->hash = nhash;
517 clhash->hashsize = nsize;
518 clhash->hashmask = nmask;
519 sch_tree_unlock(sch);
521 qdisc_class_hash_free(ohash, osize);
523 EXPORT_SYMBOL(qdisc_class_hash_grow);
525 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
527 unsigned int size = 4;
529 clhash->hash = qdisc_class_hash_alloc(size);
530 if (clhash->hash == NULL)
532 clhash->hashsize = size;
533 clhash->hashmask = size - 1;
534 clhash->hashelems = 0;
537 EXPORT_SYMBOL(qdisc_class_hash_init);
539 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
541 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
543 EXPORT_SYMBOL(qdisc_class_hash_destroy);
545 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
546 struct Qdisc_class_common *cl)
550 INIT_HLIST_NODE(&cl->hnode);
551 h = qdisc_class_hash(cl->classid, clhash->hashmask);
552 hlist_add_head(&cl->hnode, &clhash->hash[h]);
555 EXPORT_SYMBOL(qdisc_class_hash_insert);
557 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
558 struct Qdisc_class_common *cl)
560 hlist_del(&cl->hnode);
563 EXPORT_SYMBOL(qdisc_class_hash_remove);
565 /* Allocate an unique handle from space managed by kernel */
567 static u32 qdisc_alloc_handle(struct net_device *dev)
570 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
573 autohandle += TC_H_MAKE(0x10000U, 0);
574 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
575 autohandle = TC_H_MAKE(0x80000000U, 0);
576 } while (qdisc_lookup(dev, autohandle) && --i > 0);
578 return i>0 ? autohandle : 0;
581 /* Attach toplevel qdisc to device queue. */
583 static struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
586 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
587 spinlock_t *root_lock;
589 root_lock = qdisc_root_lock(oqdisc);
590 spin_lock_bh(root_lock);
592 /* Prune old scheduler */
593 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
596 /* ... and graft new one */
599 dev_queue->qdisc_sleeping = qdisc;
600 dev_queue->qdisc = &noop_qdisc;
602 spin_unlock_bh(root_lock);
607 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
609 const struct Qdisc_class_ops *cops;
615 while ((parentid = sch->parent)) {
616 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
619 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
621 WARN_ON(parentid != TC_H_ROOT);
624 cops = sch->ops->cl_ops;
625 if (cops->qlen_notify) {
626 cl = cops->get(sch, parentid);
627 cops->qlen_notify(sch, cl);
633 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
635 static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid,
636 struct Qdisc *old, struct Qdisc *new)
639 qdisc_notify(skb, n, clid, old, new);
645 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
648 * When appropriate send a netlink notification using 'skb'
651 * On success, destroy old qdisc.
654 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
655 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
656 struct Qdisc *new, struct Qdisc *old)
658 struct Qdisc *q = old;
661 if (parent == NULL) {
662 unsigned int i, num_q, ingress;
665 num_q = dev->num_tx_queues;
666 if ((q && q->flags & TCQ_F_INGRESS) ||
667 (new && new->flags & TCQ_F_INGRESS)) {
672 if (dev->flags & IFF_UP)
675 for (i = 0; i < num_q; i++) {
676 struct netdev_queue *dev_queue = &dev->rx_queue;
679 dev_queue = netdev_get_tx_queue(dev, i);
681 old = dev_graft_qdisc(dev_queue, new);
683 atomic_inc(&new->refcnt);
685 notify_and_destroy(skb, n, classid, old, new);
688 if (dev->flags & IFF_UP)
691 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
696 unsigned long cl = cops->get(parent, classid);
698 err = cops->graft(parent, cl, new, &old);
699 cops->put(parent, cl);
703 notify_and_destroy(skb, n, classid, old, new);
708 /* lockdep annotation is needed for ingress; egress gets it only for name */
709 static struct lock_class_key qdisc_tx_lock;
710 static struct lock_class_key qdisc_rx_lock;
713 Allocate and initialize new qdisc.
715 Parameters are passed via opt.
718 static struct Qdisc *
719 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
720 u32 parent, u32 handle, struct nlattr **tca, int *errp)
723 struct nlattr *kind = tca[TCA_KIND];
725 struct Qdisc_ops *ops;
726 struct qdisc_size_table *stab;
728 ops = qdisc_lookup_ops(kind);
730 if (ops == NULL && kind != NULL) {
732 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
733 /* We dropped the RTNL semaphore in order to
734 * perform the module load. So, even if we
735 * succeeded in loading the module we have to
736 * tell the caller to replay the request. We
737 * indicate this using -EAGAIN.
738 * We replay the request because the device may
739 * go away in the mean time.
742 request_module("sch_%s", name);
744 ops = qdisc_lookup_ops(kind);
746 /* We will try again qdisc_lookup_ops,
747 * so don't keep a reference.
749 module_put(ops->owner);
761 sch = qdisc_alloc(dev_queue, ops);
767 sch->parent = parent;
769 if (handle == TC_H_INGRESS) {
770 sch->flags |= TCQ_F_INGRESS;
771 handle = TC_H_MAKE(TC_H_INGRESS, 0);
772 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
775 handle = qdisc_alloc_handle(dev);
780 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
783 sch->handle = handle;
785 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
787 stab = qdisc_get_stab(tca[TCA_STAB]);
795 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
796 qdisc_root_lock(sch),
800 * Any broken qdiscs that would require
801 * a ops->reset() here? The qdisc was never
802 * in action so it shouldn't be necessary.
809 if ((parent != TC_H_ROOT) && !(sch->flags & TCQ_F_INGRESS))
810 list_add_tail(&sch->list, &dev_queue->qdisc_sleeping->list);
815 qdisc_put_stab(sch->stab);
817 kfree((char *) sch - sch->padded);
819 module_put(ops->owner);
825 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
827 struct qdisc_size_table *stab = NULL;
830 if (tca[TCA_OPTIONS]) {
831 if (sch->ops->change == NULL)
833 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
839 stab = qdisc_get_stab(tca[TCA_STAB]);
841 return PTR_ERR(stab);
844 qdisc_put_stab(sch->stab);
848 gen_replace_estimator(&sch->bstats, &sch->rate_est,
849 qdisc_root_lock(sch), tca[TCA_RATE]);
853 struct check_loop_arg
855 struct qdisc_walker w;
860 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
862 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
864 struct check_loop_arg arg;
866 if (q->ops->cl_ops == NULL)
869 arg.w.stop = arg.w.skip = arg.w.count = 0;
870 arg.w.fn = check_loop_fn;
873 q->ops->cl_ops->walk(q, &arg.w);
874 return arg.w.stop ? -ELOOP : 0;
878 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
881 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
882 struct check_loop_arg *arg = (struct check_loop_arg *)w;
884 leaf = cops->leaf(q, cl);
886 if (leaf == arg->p || arg->depth > 7)
888 return check_loop(leaf, arg->p, arg->depth + 1);
897 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
899 struct net *net = sock_net(skb->sk);
900 struct tcmsg *tcm = NLMSG_DATA(n);
901 struct nlattr *tca[TCA_MAX + 1];
902 struct net_device *dev;
903 u32 clid = tcm->tcm_parent;
904 struct Qdisc *q = NULL;
905 struct Qdisc *p = NULL;
908 if (net != &init_net)
911 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
914 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
919 if (clid != TC_H_ROOT) {
920 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
921 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
923 q = qdisc_leaf(p, clid);
924 } else { /* ingress */
925 q = dev->rx_queue.qdisc_sleeping;
928 struct netdev_queue *dev_queue;
929 dev_queue = netdev_get_tx_queue(dev, 0);
930 q = dev_queue->qdisc_sleeping;
935 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
938 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
942 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
945 if (n->nlmsg_type == RTM_DELQDISC) {
950 if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0)
953 qdisc_notify(skb, n, clid, NULL, q);
962 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
964 struct net *net = sock_net(skb->sk);
966 struct nlattr *tca[TCA_MAX + 1];
967 struct net_device *dev;
972 if (net != &init_net)
976 /* Reinit, just in case something touches this. */
978 clid = tcm->tcm_parent;
981 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
984 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
989 if (clid != TC_H_ROOT) {
990 if (clid != TC_H_INGRESS) {
991 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
993 q = qdisc_leaf(p, clid);
994 } else { /*ingress */
995 q = dev->rx_queue.qdisc_sleeping;
998 struct netdev_queue *dev_queue;
999 dev_queue = netdev_get_tx_queue(dev, 0);
1000 q = dev_queue->qdisc_sleeping;
1003 /* It may be default qdisc, ignore it */
1004 if (q && q->handle == 0)
1007 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1008 if (tcm->tcm_handle) {
1009 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
1011 if (TC_H_MIN(tcm->tcm_handle))
1013 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
1014 goto create_n_graft;
1015 if (n->nlmsg_flags&NLM_F_EXCL)
1017 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1020 (p && check_loop(q, p, 0)))
1022 atomic_inc(&q->refcnt);
1026 goto create_n_graft;
1028 /* This magic test requires explanation.
1030 * We know, that some child q is already
1031 * attached to this parent and have choice:
1032 * either to change it or to create/graft new one.
1034 * 1. We are allowed to create/graft only
1035 * if CREATE and REPLACE flags are set.
1037 * 2. If EXCL is set, requestor wanted to say,
1038 * that qdisc tcm_handle is not expected
1039 * to exist, so that we choose create/graft too.
1041 * 3. The last case is when no flags are set.
1042 * Alas, it is sort of hole in API, we
1043 * cannot decide what to do unambiguously.
1044 * For now we select create/graft, if
1045 * user gave KIND, which does not match existing.
1047 if ((n->nlmsg_flags&NLM_F_CREATE) &&
1048 (n->nlmsg_flags&NLM_F_REPLACE) &&
1049 ((n->nlmsg_flags&NLM_F_EXCL) ||
1051 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1052 goto create_n_graft;
1056 if (!tcm->tcm_handle)
1058 q = qdisc_lookup(dev, tcm->tcm_handle);
1061 /* Change qdisc parameters */
1064 if (n->nlmsg_flags&NLM_F_EXCL)
1066 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1068 err = qdisc_change(q, tca);
1070 qdisc_notify(skb, n, clid, NULL, q);
1074 if (!(n->nlmsg_flags&NLM_F_CREATE))
1076 if (clid == TC_H_INGRESS)
1077 q = qdisc_create(dev, &dev->rx_queue,
1078 tcm->tcm_parent, tcm->tcm_parent,
1081 q = qdisc_create(dev, netdev_get_tx_queue(dev, 0),
1082 tcm->tcm_parent, tcm->tcm_handle,
1092 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1102 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1103 u32 pid, u32 seq, u16 flags, int event)
1106 struct nlmsghdr *nlh;
1107 unsigned char *b = skb_tail_pointer(skb);
1110 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1111 tcm = NLMSG_DATA(nlh);
1112 tcm->tcm_family = AF_UNSPEC;
1115 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1116 tcm->tcm_parent = clid;
1117 tcm->tcm_handle = q->handle;
1118 tcm->tcm_info = atomic_read(&q->refcnt);
1119 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1120 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1121 goto nla_put_failure;
1122 q->qstats.qlen = q->q.qlen;
1124 if (q->stab && qdisc_dump_stab(skb, q->stab) < 0)
1125 goto nla_put_failure;
1127 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1128 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
1129 goto nla_put_failure;
1131 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1132 goto nla_put_failure;
1134 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1135 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
1136 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1137 goto nla_put_failure;
1139 if (gnet_stats_finish_copy(&d) < 0)
1140 goto nla_put_failure;
1142 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1151 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1152 u32 clid, struct Qdisc *old, struct Qdisc *new)
1154 struct sk_buff *skb;
1155 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1157 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1161 if (old && old->handle) {
1162 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
1166 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1171 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1178 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1180 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1183 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1184 struct netlink_callback *cb,
1185 int *q_idx_p, int s_q_idx)
1187 int ret = 0, q_idx = *q_idx_p;
1194 if (q_idx < s_q_idx) {
1197 if (!tc_qdisc_dump_ignore(q) &&
1198 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1199 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1203 list_for_each_entry(q, &root->list, list) {
1204 if (q_idx < s_q_idx) {
1208 if (!tc_qdisc_dump_ignore(q) &&
1209 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1210 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1223 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1225 struct net *net = sock_net(skb->sk);
1228 struct net_device *dev;
1230 if (net != &init_net)
1233 s_idx = cb->args[0];
1234 s_q_idx = q_idx = cb->args[1];
1235 read_lock(&dev_base_lock);
1237 for_each_netdev(&init_net, dev) {
1238 struct netdev_queue *dev_queue;
1246 dev_queue = netdev_get_tx_queue(dev, 0);
1247 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1250 dev_queue = &dev->rx_queue;
1251 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1259 read_unlock(&dev_base_lock);
1262 cb->args[1] = q_idx;
1269 /************************************************
1270 * Traffic classes manipulation. *
1271 ************************************************/
1275 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1277 struct net *net = sock_net(skb->sk);
1278 struct netdev_queue *dev_queue;
1279 struct tcmsg *tcm = NLMSG_DATA(n);
1280 struct nlattr *tca[TCA_MAX + 1];
1281 struct net_device *dev;
1282 struct Qdisc *q = NULL;
1283 const struct Qdisc_class_ops *cops;
1284 unsigned long cl = 0;
1285 unsigned long new_cl;
1286 u32 pid = tcm->tcm_parent;
1287 u32 clid = tcm->tcm_handle;
1288 u32 qid = TC_H_MAJ(clid);
1291 if (net != &init_net)
1294 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1297 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1302 parent == TC_H_UNSPEC - unspecified parent.
1303 parent == TC_H_ROOT - class is root, which has no parent.
1304 parent == X:0 - parent is root class.
1305 parent == X:Y - parent is a node in hierarchy.
1306 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1308 handle == 0:0 - generate handle from kernel pool.
1309 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1310 handle == X:Y - clear.
1311 handle == X:0 - root class.
1314 /* Step 1. Determine qdisc handle X:0 */
1316 dev_queue = netdev_get_tx_queue(dev, 0);
1317 if (pid != TC_H_ROOT) {
1318 u32 qid1 = TC_H_MAJ(pid);
1321 /* If both majors are known, they must be identical. */
1326 } else if (qid == 0)
1327 qid = dev_queue->qdisc_sleeping->handle;
1329 /* Now qid is genuine qdisc handle consistent
1330 both with parent and child.
1332 TC_H_MAJ(pid) still may be unspecified, complete it now.
1335 pid = TC_H_MAKE(qid, pid);
1338 qid = dev_queue->qdisc_sleeping->handle;
1341 /* OK. Locate qdisc */
1342 if ((q = qdisc_lookup(dev, qid)) == NULL)
1345 /* An check that it supports classes */
1346 cops = q->ops->cl_ops;
1350 /* Now try to get class */
1352 if (pid == TC_H_ROOT)
1355 clid = TC_H_MAKE(qid, clid);
1358 cl = cops->get(q, clid);
1362 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1365 switch (n->nlmsg_type) {
1368 if (n->nlmsg_flags&NLM_F_EXCL)
1372 err = cops->delete(q, cl);
1374 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1377 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1386 err = cops->change(q, clid, pid, tca, &new_cl);
1388 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1398 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1400 u32 pid, u32 seq, u16 flags, int event)
1403 struct nlmsghdr *nlh;
1404 unsigned char *b = skb_tail_pointer(skb);
1406 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1408 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1409 tcm = NLMSG_DATA(nlh);
1410 tcm->tcm_family = AF_UNSPEC;
1411 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1412 tcm->tcm_parent = q->handle;
1413 tcm->tcm_handle = q->handle;
1415 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1416 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1417 goto nla_put_failure;
1419 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1420 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
1421 goto nla_put_failure;
1423 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1424 goto nla_put_failure;
1426 if (gnet_stats_finish_copy(&d) < 0)
1427 goto nla_put_failure;
1429 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1438 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1439 struct Qdisc *q, unsigned long cl, int event)
1441 struct sk_buff *skb;
1442 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1444 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1448 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1453 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1456 struct qdisc_dump_args
1458 struct qdisc_walker w;
1459 struct sk_buff *skb;
1460 struct netlink_callback *cb;
1463 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1465 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1467 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1468 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1471 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1472 struct tcmsg *tcm, struct netlink_callback *cb,
1475 struct qdisc_dump_args arg;
1477 if (tc_qdisc_dump_ignore(q) ||
1478 *t_p < s_t || !q->ops->cl_ops ||
1480 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1485 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1486 arg.w.fn = qdisc_class_dump;
1490 arg.w.skip = cb->args[1];
1492 q->ops->cl_ops->walk(q, &arg.w);
1493 cb->args[1] = arg.w.count;
1500 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1501 struct tcmsg *tcm, struct netlink_callback *cb,
1509 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1512 list_for_each_entry(q, &root->list, list) {
1513 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1520 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1522 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1523 struct net *net = sock_net(skb->sk);
1524 struct netdev_queue *dev_queue;
1525 struct net_device *dev;
1528 if (net != &init_net)
1531 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1533 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1539 dev_queue = netdev_get_tx_queue(dev, 0);
1540 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1543 dev_queue = &dev->rx_queue;
1544 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1554 /* Main classifier routine: scans classifier chain attached
1555 to this qdisc, (optionally) tests for protocol and asks
1556 specific classifiers.
1558 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1559 struct tcf_result *res)
1561 __be16 protocol = skb->protocol;
1564 for (; tp; tp = tp->next) {
1565 if ((tp->protocol == protocol ||
1566 tp->protocol == htons(ETH_P_ALL)) &&
1567 (err = tp->classify(skb, tp, res)) >= 0) {
1568 #ifdef CONFIG_NET_CLS_ACT
1569 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1570 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1577 EXPORT_SYMBOL(tc_classify_compat);
1579 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1580 struct tcf_result *res)
1584 #ifdef CONFIG_NET_CLS_ACT
1585 struct tcf_proto *otp = tp;
1588 protocol = skb->protocol;
1590 err = tc_classify_compat(skb, tp, res);
1591 #ifdef CONFIG_NET_CLS_ACT
1592 if (err == TC_ACT_RECLASSIFY) {
1593 u32 verd = G_TC_VERD(skb->tc_verd);
1596 if (verd++ >= MAX_REC_LOOP) {
1597 printk("rule prio %u protocol %02x reclassify loop, "
1599 tp->prio&0xffff, ntohs(tp->protocol));
1602 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1608 EXPORT_SYMBOL(tc_classify);
1610 void tcf_destroy(struct tcf_proto *tp)
1612 tp->ops->destroy(tp);
1613 module_put(tp->ops->owner);
1617 void tcf_destroy_chain(struct tcf_proto **fl)
1619 struct tcf_proto *tp;
1621 while ((tp = *fl) != NULL) {
1626 EXPORT_SYMBOL(tcf_destroy_chain);
1628 #ifdef CONFIG_PROC_FS
1629 static int psched_show(struct seq_file *seq, void *v)
1633 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1634 seq_printf(seq, "%08x %08x %08x %08x\n",
1635 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1637 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1642 static int psched_open(struct inode *inode, struct file *file)
1644 return single_open(file, psched_show, PDE(inode)->data);
1647 static const struct file_operations psched_fops = {
1648 .owner = THIS_MODULE,
1649 .open = psched_open,
1651 .llseek = seq_lseek,
1652 .release = single_release,
1656 static int __init pktsched_init(void)
1658 register_qdisc(&pfifo_qdisc_ops);
1659 register_qdisc(&bfifo_qdisc_ops);
1660 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1662 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1663 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1664 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1665 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1666 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1667 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1672 subsys_initcall(pktsched_init);