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);
644 sch_tree_unlock(old);
648 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
651 * When appropriate send a netlink notification using 'skb'
654 * On success, destroy old qdisc.
657 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
658 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
659 struct Qdisc *new, struct Qdisc *old)
661 struct Qdisc *q = old;
664 if (parent == NULL) {
665 unsigned int i, num_q, ingress;
668 num_q = dev->num_tx_queues;
669 if ((q && q->flags & TCQ_F_INGRESS) ||
670 (new && new->flags & TCQ_F_INGRESS)) {
675 if (dev->flags & IFF_UP)
678 for (i = 0; i < num_q; i++) {
679 struct netdev_queue *dev_queue = &dev->rx_queue;
682 dev_queue = netdev_get_tx_queue(dev, i);
684 old = dev_graft_qdisc(dev_queue, new);
686 atomic_inc(&new->refcnt);
688 notify_and_destroy(skb, n, classid, old, new);
691 if (dev->flags & IFF_UP)
694 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
699 unsigned long cl = cops->get(parent, classid);
701 err = cops->graft(parent, cl, new, &old);
702 cops->put(parent, cl);
706 notify_and_destroy(skb, n, classid, old, new);
711 /* lockdep annotation is needed for ingress; egress gets it only for name */
712 static struct lock_class_key qdisc_tx_lock;
713 static struct lock_class_key qdisc_rx_lock;
716 Allocate and initialize new qdisc.
718 Parameters are passed via opt.
721 static struct Qdisc *
722 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
723 u32 parent, u32 handle, struct nlattr **tca, int *errp)
726 struct nlattr *kind = tca[TCA_KIND];
728 struct Qdisc_ops *ops;
729 struct qdisc_size_table *stab;
731 ops = qdisc_lookup_ops(kind);
733 if (ops == NULL && kind != NULL) {
735 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
736 /* We dropped the RTNL semaphore in order to
737 * perform the module load. So, even if we
738 * succeeded in loading the module we have to
739 * tell the caller to replay the request. We
740 * indicate this using -EAGAIN.
741 * We replay the request because the device may
742 * go away in the mean time.
745 request_module("sch_%s", name);
747 ops = qdisc_lookup_ops(kind);
749 /* We will try again qdisc_lookup_ops,
750 * so don't keep a reference.
752 module_put(ops->owner);
764 sch = qdisc_alloc(dev_queue, ops);
770 sch->parent = parent;
772 if (handle == TC_H_INGRESS) {
773 sch->flags |= TCQ_F_INGRESS;
774 handle = TC_H_MAKE(TC_H_INGRESS, 0);
775 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
778 handle = qdisc_alloc_handle(dev);
783 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
786 sch->handle = handle;
788 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
790 stab = qdisc_get_stab(tca[TCA_STAB]);
798 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
799 qdisc_root_lock(sch),
803 * Any broken qdiscs that would require
804 * a ops->reset() here? The qdisc was never
805 * in action so it shouldn't be necessary.
812 if ((parent != TC_H_ROOT) && !(sch->flags & TCQ_F_INGRESS))
813 list_add_tail(&sch->list, &dev_queue->qdisc_sleeping->list);
818 qdisc_put_stab(sch->stab);
820 kfree((char *) sch - sch->padded);
822 module_put(ops->owner);
828 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
830 struct qdisc_size_table *stab = NULL;
833 if (tca[TCA_OPTIONS]) {
834 if (sch->ops->change == NULL)
836 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
842 stab = qdisc_get_stab(tca[TCA_STAB]);
844 return PTR_ERR(stab);
847 qdisc_put_stab(sch->stab);
851 gen_replace_estimator(&sch->bstats, &sch->rate_est,
852 qdisc_root_lock(sch), tca[TCA_RATE]);
856 struct check_loop_arg
858 struct qdisc_walker w;
863 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
865 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
867 struct check_loop_arg arg;
869 if (q->ops->cl_ops == NULL)
872 arg.w.stop = arg.w.skip = arg.w.count = 0;
873 arg.w.fn = check_loop_fn;
876 q->ops->cl_ops->walk(q, &arg.w);
877 return arg.w.stop ? -ELOOP : 0;
881 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
884 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
885 struct check_loop_arg *arg = (struct check_loop_arg *)w;
887 leaf = cops->leaf(q, cl);
889 if (leaf == arg->p || arg->depth > 7)
891 return check_loop(leaf, arg->p, arg->depth + 1);
900 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
902 struct net *net = sock_net(skb->sk);
903 struct tcmsg *tcm = NLMSG_DATA(n);
904 struct nlattr *tca[TCA_MAX + 1];
905 struct net_device *dev;
906 u32 clid = tcm->tcm_parent;
907 struct Qdisc *q = NULL;
908 struct Qdisc *p = NULL;
911 if (net != &init_net)
914 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
917 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
922 if (clid != TC_H_ROOT) {
923 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
924 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
926 q = qdisc_leaf(p, clid);
927 } else { /* ingress */
928 q = dev->rx_queue.qdisc_sleeping;
931 struct netdev_queue *dev_queue;
932 dev_queue = netdev_get_tx_queue(dev, 0);
933 q = dev_queue->qdisc_sleeping;
938 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
941 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
945 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
948 if (n->nlmsg_type == RTM_DELQDISC) {
953 if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0)
956 qdisc_notify(skb, n, clid, NULL, q);
965 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
967 struct net *net = sock_net(skb->sk);
969 struct nlattr *tca[TCA_MAX + 1];
970 struct net_device *dev;
975 if (net != &init_net)
979 /* Reinit, just in case something touches this. */
981 clid = tcm->tcm_parent;
984 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
987 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
992 if (clid != TC_H_ROOT) {
993 if (clid != TC_H_INGRESS) {
994 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
996 q = qdisc_leaf(p, clid);
997 } else { /*ingress */
998 q = dev->rx_queue.qdisc_sleeping;
1001 struct netdev_queue *dev_queue;
1002 dev_queue = netdev_get_tx_queue(dev, 0);
1003 q = dev_queue->qdisc_sleeping;
1006 /* It may be default qdisc, ignore it */
1007 if (q && q->handle == 0)
1010 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1011 if (tcm->tcm_handle) {
1012 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
1014 if (TC_H_MIN(tcm->tcm_handle))
1016 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
1017 goto create_n_graft;
1018 if (n->nlmsg_flags&NLM_F_EXCL)
1020 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1023 (p && check_loop(q, p, 0)))
1025 atomic_inc(&q->refcnt);
1029 goto create_n_graft;
1031 /* This magic test requires explanation.
1033 * We know, that some child q is already
1034 * attached to this parent and have choice:
1035 * either to change it or to create/graft new one.
1037 * 1. We are allowed to create/graft only
1038 * if CREATE and REPLACE flags are set.
1040 * 2. If EXCL is set, requestor wanted to say,
1041 * that qdisc tcm_handle is not expected
1042 * to exist, so that we choose create/graft too.
1044 * 3. The last case is when no flags are set.
1045 * Alas, it is sort of hole in API, we
1046 * cannot decide what to do unambiguously.
1047 * For now we select create/graft, if
1048 * user gave KIND, which does not match existing.
1050 if ((n->nlmsg_flags&NLM_F_CREATE) &&
1051 (n->nlmsg_flags&NLM_F_REPLACE) &&
1052 ((n->nlmsg_flags&NLM_F_EXCL) ||
1054 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1055 goto create_n_graft;
1059 if (!tcm->tcm_handle)
1061 q = qdisc_lookup(dev, tcm->tcm_handle);
1064 /* Change qdisc parameters */
1067 if (n->nlmsg_flags&NLM_F_EXCL)
1069 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1071 err = qdisc_change(q, tca);
1073 qdisc_notify(skb, n, clid, NULL, q);
1077 if (!(n->nlmsg_flags&NLM_F_CREATE))
1079 if (clid == TC_H_INGRESS)
1080 q = qdisc_create(dev, &dev->rx_queue,
1081 tcm->tcm_parent, tcm->tcm_parent,
1084 q = qdisc_create(dev, netdev_get_tx_queue(dev, 0),
1085 tcm->tcm_parent, tcm->tcm_handle,
1095 spinlock_t *root_lock;
1097 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1100 root_lock = qdisc_root_lock(q);
1101 spin_lock_bh(root_lock);
1103 spin_unlock_bh(root_lock);
1111 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1112 u32 pid, u32 seq, u16 flags, int event)
1115 struct nlmsghdr *nlh;
1116 unsigned char *b = skb_tail_pointer(skb);
1119 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1120 tcm = NLMSG_DATA(nlh);
1121 tcm->tcm_family = AF_UNSPEC;
1124 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1125 tcm->tcm_parent = clid;
1126 tcm->tcm_handle = q->handle;
1127 tcm->tcm_info = atomic_read(&q->refcnt);
1128 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1129 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1130 goto nla_put_failure;
1131 q->qstats.qlen = q->q.qlen;
1133 if (q->stab && qdisc_dump_stab(skb, q->stab) < 0)
1134 goto nla_put_failure;
1136 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1137 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
1138 goto nla_put_failure;
1140 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1141 goto nla_put_failure;
1143 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1144 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
1145 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1146 goto nla_put_failure;
1148 if (gnet_stats_finish_copy(&d) < 0)
1149 goto nla_put_failure;
1151 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1160 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1161 u32 clid, struct Qdisc *old, struct Qdisc *new)
1163 struct sk_buff *skb;
1164 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1166 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1170 if (old && old->handle) {
1171 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
1175 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1180 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1187 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1189 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1192 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1193 struct netlink_callback *cb,
1194 int *q_idx_p, int s_q_idx)
1196 int ret = 0, q_idx = *q_idx_p;
1203 if (q_idx < s_q_idx) {
1206 if (!tc_qdisc_dump_ignore(q) &&
1207 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1208 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1212 list_for_each_entry(q, &root->list, list) {
1213 if (q_idx < s_q_idx) {
1217 if (!tc_qdisc_dump_ignore(q) &&
1218 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1219 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1232 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1234 struct net *net = sock_net(skb->sk);
1237 struct net_device *dev;
1239 if (net != &init_net)
1242 s_idx = cb->args[0];
1243 s_q_idx = q_idx = cb->args[1];
1244 read_lock(&dev_base_lock);
1246 for_each_netdev(&init_net, dev) {
1247 struct netdev_queue *dev_queue;
1255 dev_queue = netdev_get_tx_queue(dev, 0);
1256 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1259 dev_queue = &dev->rx_queue;
1260 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1268 read_unlock(&dev_base_lock);
1271 cb->args[1] = q_idx;
1278 /************************************************
1279 * Traffic classes manipulation. *
1280 ************************************************/
1284 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1286 struct net *net = sock_net(skb->sk);
1287 struct netdev_queue *dev_queue;
1288 struct tcmsg *tcm = NLMSG_DATA(n);
1289 struct nlattr *tca[TCA_MAX + 1];
1290 struct net_device *dev;
1291 struct Qdisc *q = NULL;
1292 const struct Qdisc_class_ops *cops;
1293 unsigned long cl = 0;
1294 unsigned long new_cl;
1295 u32 pid = tcm->tcm_parent;
1296 u32 clid = tcm->tcm_handle;
1297 u32 qid = TC_H_MAJ(clid);
1300 if (net != &init_net)
1303 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1306 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1311 parent == TC_H_UNSPEC - unspecified parent.
1312 parent == TC_H_ROOT - class is root, which has no parent.
1313 parent == X:0 - parent is root class.
1314 parent == X:Y - parent is a node in hierarchy.
1315 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1317 handle == 0:0 - generate handle from kernel pool.
1318 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1319 handle == X:Y - clear.
1320 handle == X:0 - root class.
1323 /* Step 1. Determine qdisc handle X:0 */
1325 dev_queue = netdev_get_tx_queue(dev, 0);
1326 if (pid != TC_H_ROOT) {
1327 u32 qid1 = TC_H_MAJ(pid);
1330 /* If both majors are known, they must be identical. */
1335 } else if (qid == 0)
1336 qid = dev_queue->qdisc_sleeping->handle;
1338 /* Now qid is genuine qdisc handle consistent
1339 both with parent and child.
1341 TC_H_MAJ(pid) still may be unspecified, complete it now.
1344 pid = TC_H_MAKE(qid, pid);
1347 qid = dev_queue->qdisc_sleeping->handle;
1350 /* OK. Locate qdisc */
1351 if ((q = qdisc_lookup(dev, qid)) == NULL)
1354 /* An check that it supports classes */
1355 cops = q->ops->cl_ops;
1359 /* Now try to get class */
1361 if (pid == TC_H_ROOT)
1364 clid = TC_H_MAKE(qid, clid);
1367 cl = cops->get(q, clid);
1371 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1374 switch (n->nlmsg_type) {
1377 if (n->nlmsg_flags&NLM_F_EXCL)
1381 err = cops->delete(q, cl);
1383 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1386 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1395 err = cops->change(q, clid, pid, tca, &new_cl);
1397 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1407 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1409 u32 pid, u32 seq, u16 flags, int event)
1412 struct nlmsghdr *nlh;
1413 unsigned char *b = skb_tail_pointer(skb);
1415 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1417 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1418 tcm = NLMSG_DATA(nlh);
1419 tcm->tcm_family = AF_UNSPEC;
1420 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1421 tcm->tcm_parent = q->handle;
1422 tcm->tcm_handle = q->handle;
1424 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1425 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1426 goto nla_put_failure;
1428 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1429 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
1430 goto nla_put_failure;
1432 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1433 goto nla_put_failure;
1435 if (gnet_stats_finish_copy(&d) < 0)
1436 goto nla_put_failure;
1438 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1447 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1448 struct Qdisc *q, unsigned long cl, int event)
1450 struct sk_buff *skb;
1451 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1453 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1457 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1462 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1465 struct qdisc_dump_args
1467 struct qdisc_walker w;
1468 struct sk_buff *skb;
1469 struct netlink_callback *cb;
1472 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1474 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1476 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1477 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1480 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1481 struct tcmsg *tcm, struct netlink_callback *cb,
1484 struct qdisc_dump_args arg;
1486 if (tc_qdisc_dump_ignore(q) ||
1487 *t_p < s_t || !q->ops->cl_ops ||
1489 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1494 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1495 arg.w.fn = qdisc_class_dump;
1499 arg.w.skip = cb->args[1];
1501 q->ops->cl_ops->walk(q, &arg.w);
1502 cb->args[1] = arg.w.count;
1509 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1510 struct tcmsg *tcm, struct netlink_callback *cb,
1518 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1521 list_for_each_entry(q, &root->list, list) {
1522 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1529 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1531 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1532 struct net *net = sock_net(skb->sk);
1533 struct netdev_queue *dev_queue;
1534 struct net_device *dev;
1537 if (net != &init_net)
1540 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1542 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1548 dev_queue = netdev_get_tx_queue(dev, 0);
1549 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1552 dev_queue = &dev->rx_queue;
1553 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1563 /* Main classifier routine: scans classifier chain attached
1564 to this qdisc, (optionally) tests for protocol and asks
1565 specific classifiers.
1567 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1568 struct tcf_result *res)
1570 __be16 protocol = skb->protocol;
1573 for (; tp; tp = tp->next) {
1574 if ((tp->protocol == protocol ||
1575 tp->protocol == htons(ETH_P_ALL)) &&
1576 (err = tp->classify(skb, tp, res)) >= 0) {
1577 #ifdef CONFIG_NET_CLS_ACT
1578 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1579 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1586 EXPORT_SYMBOL(tc_classify_compat);
1588 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1589 struct tcf_result *res)
1593 #ifdef CONFIG_NET_CLS_ACT
1594 struct tcf_proto *otp = tp;
1597 protocol = skb->protocol;
1599 err = tc_classify_compat(skb, tp, res);
1600 #ifdef CONFIG_NET_CLS_ACT
1601 if (err == TC_ACT_RECLASSIFY) {
1602 u32 verd = G_TC_VERD(skb->tc_verd);
1605 if (verd++ >= MAX_REC_LOOP) {
1606 printk("rule prio %u protocol %02x reclassify loop, "
1608 tp->prio&0xffff, ntohs(tp->protocol));
1611 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1617 EXPORT_SYMBOL(tc_classify);
1619 void tcf_destroy(struct tcf_proto *tp)
1621 tp->ops->destroy(tp);
1622 module_put(tp->ops->owner);
1626 void tcf_destroy_chain(struct tcf_proto **fl)
1628 struct tcf_proto *tp;
1630 while ((tp = *fl) != NULL) {
1635 EXPORT_SYMBOL(tcf_destroy_chain);
1637 #ifdef CONFIG_PROC_FS
1638 static int psched_show(struct seq_file *seq, void *v)
1642 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1643 seq_printf(seq, "%08x %08x %08x %08x\n",
1644 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1646 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1651 static int psched_open(struct inode *inode, struct file *file)
1653 return single_open(file, psched_show, PDE(inode)->data);
1656 static const struct file_operations psched_fops = {
1657 .owner = THIS_MODULE,
1658 .open = psched_open,
1660 .llseek = seq_lseek,
1661 .release = single_release,
1665 static int __init pktsched_init(void)
1667 register_qdisc(&pfifo_qdisc_ops);
1668 register_qdisc(&bfifo_qdisc_ops);
1669 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1671 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1672 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1673 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1674 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1675 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1676 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1681 subsys_initcall(pktsched_init);