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>
31 #include <linux/slab.h>
33 #include <net/net_namespace.h>
35 #include <net/netlink.h>
36 #include <net/pkt_sched.h>
38 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
39 struct nlmsghdr *n, u32 clid,
40 struct Qdisc *old, struct Qdisc *new);
41 static int tclass_notify(struct net *net, struct sk_buff *oskb,
42 struct nlmsghdr *n, struct Qdisc *q,
43 unsigned long cl, int event);
50 This file consists of two interrelated parts:
52 1. queueing disciplines manager frontend.
53 2. traffic classes manager frontend.
55 Generally, queueing discipline ("qdisc") is a black box,
56 which is able to enqueue packets and to dequeue them (when
57 device is ready to send something) in order and at times
58 determined by algorithm hidden in it.
60 qdisc's are divided to two categories:
61 - "queues", which have no internal structure visible from outside.
62 - "schedulers", which split all the packets to "traffic classes",
63 using "packet classifiers" (look at cls_api.c)
65 In turn, classes may have child qdiscs (as rule, queues)
66 attached to them etc. etc. etc.
68 The goal of the routines in this file is to translate
69 information supplied by user in the form of handles
70 to more intelligible for kernel form, to make some sanity
71 checks and part of work, which is common to all qdiscs
72 and to provide rtnetlink notifications.
74 All real intelligent work is done inside qdisc modules.
78 Every discipline has two major routines: enqueue and dequeue.
82 dequeue usually returns a skb to send. It is allowed to return NULL,
83 but it does not mean that queue is empty, it just means that
84 discipline does not want to send anything this time.
85 Queue is really empty if q->q.qlen == 0.
86 For complicated disciplines with multiple queues q->q is not
87 real packet queue, but however q->q.qlen must be valid.
91 enqueue returns 0, if packet was enqueued successfully.
92 If packet (this one or another one) was dropped, it returns
94 NET_XMIT_DROP - this packet dropped
95 Expected action: do not backoff, but wait until queue will clear.
96 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
97 Expected action: backoff or ignore
98 NET_XMIT_POLICED - dropped by police.
99 Expected action: backoff or error to real-time apps.
105 like dequeue but without removing a packet from the queue
109 returns qdisc to initial state: purge all buffers, clear all
110 timers, counters (except for statistics) etc.
114 initializes newly created qdisc.
118 destroys resources allocated by init and during lifetime of qdisc.
122 changes qdisc parameters.
125 /* Protects list of registered TC modules. It is pure SMP lock. */
126 static DEFINE_RWLOCK(qdisc_mod_lock);
129 /************************************************
130 * Queueing disciplines manipulation. *
131 ************************************************/
134 /* The list of all installed queueing disciplines. */
136 static struct Qdisc_ops *qdisc_base;
138 /* Register/uregister queueing discipline */
140 int register_qdisc(struct Qdisc_ops *qops)
142 struct Qdisc_ops *q, **qp;
145 write_lock(&qdisc_mod_lock);
146 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
147 if (!strcmp(qops->id, q->id))
150 if (qops->enqueue == NULL)
151 qops->enqueue = noop_qdisc_ops.enqueue;
152 if (qops->peek == NULL) {
153 if (qops->dequeue == NULL)
154 qops->peek = noop_qdisc_ops.peek;
158 if (qops->dequeue == NULL)
159 qops->dequeue = noop_qdisc_ops.dequeue;
162 const struct Qdisc_class_ops *cops = qops->cl_ops;
164 if (!(cops->get && cops->put && cops->walk && cops->leaf))
167 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf))
175 write_unlock(&qdisc_mod_lock);
182 EXPORT_SYMBOL(register_qdisc);
184 int unregister_qdisc(struct Qdisc_ops *qops)
186 struct Qdisc_ops *q, **qp;
189 write_lock(&qdisc_mod_lock);
190 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
198 write_unlock(&qdisc_mod_lock);
201 EXPORT_SYMBOL(unregister_qdisc);
203 /* We know handle. Find qdisc among all qdisc's attached to device
204 (root qdisc, all its children, children of children etc.)
207 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
211 if (!(root->flags & TCQ_F_BUILTIN) &&
212 root->handle == handle)
215 list_for_each_entry(q, &root->list, list) {
216 if (q->handle == handle)
222 static void qdisc_list_add(struct Qdisc *q)
224 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
225 list_add_tail(&q->list, &qdisc_dev(q)->qdisc->list);
228 void qdisc_list_del(struct Qdisc *q)
230 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
233 EXPORT_SYMBOL(qdisc_list_del);
235 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
239 q = qdisc_match_from_root(dev->qdisc, handle);
243 if (dev_ingress_queue(dev))
244 q = qdisc_match_from_root(
245 dev_ingress_queue(dev)->qdisc_sleeping,
251 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
255 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
259 cl = cops->get(p, classid);
263 leaf = cops->leaf(p, cl);
268 /* Find queueing discipline by name */
270 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
272 struct Qdisc_ops *q = NULL;
275 read_lock(&qdisc_mod_lock);
276 for (q = qdisc_base; q; q = q->next) {
277 if (nla_strcmp(kind, q->id) == 0) {
278 if (!try_module_get(q->owner))
283 read_unlock(&qdisc_mod_lock);
288 static struct qdisc_rate_table *qdisc_rtab_list;
290 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
292 struct qdisc_rate_table *rtab;
294 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
295 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
301 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
302 nla_len(tab) != TC_RTAB_SIZE)
305 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
309 memcpy(rtab->data, nla_data(tab), 1024);
310 rtab->next = qdisc_rtab_list;
311 qdisc_rtab_list = rtab;
315 EXPORT_SYMBOL(qdisc_get_rtab);
317 void qdisc_put_rtab(struct qdisc_rate_table *tab)
319 struct qdisc_rate_table *rtab, **rtabp;
321 if (!tab || --tab->refcnt)
324 for (rtabp = &qdisc_rtab_list;
325 (rtab = *rtabp) != NULL;
326 rtabp = &rtab->next) {
334 EXPORT_SYMBOL(qdisc_put_rtab);
336 static LIST_HEAD(qdisc_stab_list);
337 static DEFINE_SPINLOCK(qdisc_stab_lock);
339 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
340 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
341 [TCA_STAB_DATA] = { .type = NLA_BINARY },
344 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
346 struct nlattr *tb[TCA_STAB_MAX + 1];
347 struct qdisc_size_table *stab;
348 struct tc_sizespec *s;
349 unsigned int tsize = 0;
353 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
356 if (!tb[TCA_STAB_BASE])
357 return ERR_PTR(-EINVAL);
359 s = nla_data(tb[TCA_STAB_BASE]);
362 if (!tb[TCA_STAB_DATA])
363 return ERR_PTR(-EINVAL);
364 tab = nla_data(tb[TCA_STAB_DATA]);
365 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
368 if (tsize != s->tsize || (!tab && tsize > 0))
369 return ERR_PTR(-EINVAL);
371 spin_lock(&qdisc_stab_lock);
373 list_for_each_entry(stab, &qdisc_stab_list, list) {
374 if (memcmp(&stab->szopts, s, sizeof(*s)))
376 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
379 spin_unlock(&qdisc_stab_lock);
383 spin_unlock(&qdisc_stab_lock);
385 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
387 return ERR_PTR(-ENOMEM);
392 memcpy(stab->data, tab, tsize * sizeof(u16));
394 spin_lock(&qdisc_stab_lock);
395 list_add_tail(&stab->list, &qdisc_stab_list);
396 spin_unlock(&qdisc_stab_lock);
401 void qdisc_put_stab(struct qdisc_size_table *tab)
406 spin_lock(&qdisc_stab_lock);
408 if (--tab->refcnt == 0) {
409 list_del(&tab->list);
413 spin_unlock(&qdisc_stab_lock);
415 EXPORT_SYMBOL(qdisc_put_stab);
417 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
421 nest = nla_nest_start(skb, TCA_STAB);
423 goto nla_put_failure;
424 NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts);
425 nla_nest_end(skb, nest);
433 void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab)
437 pkt_len = skb->len + stab->szopts.overhead;
438 if (unlikely(!stab->szopts.tsize))
441 slot = pkt_len + stab->szopts.cell_align;
442 if (unlikely(slot < 0))
445 slot >>= stab->szopts.cell_log;
446 if (likely(slot < stab->szopts.tsize))
447 pkt_len = stab->data[slot];
449 pkt_len = stab->data[stab->szopts.tsize - 1] *
450 (slot / stab->szopts.tsize) +
451 stab->data[slot % stab->szopts.tsize];
453 pkt_len <<= stab->szopts.size_log;
455 if (unlikely(pkt_len < 1))
457 qdisc_skb_cb(skb)->pkt_len = pkt_len;
459 EXPORT_SYMBOL(qdisc_calculate_pkt_len);
461 void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
463 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
464 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
465 txt, qdisc->ops->id, qdisc->handle >> 16);
466 qdisc->flags |= TCQ_F_WARN_NONWC;
469 EXPORT_SYMBOL(qdisc_warn_nonwc);
471 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
473 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
476 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
477 __netif_schedule(qdisc_root(wd->qdisc));
479 return HRTIMER_NORESTART;
482 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
484 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
485 wd->timer.function = qdisc_watchdog;
488 EXPORT_SYMBOL(qdisc_watchdog_init);
490 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
494 if (test_bit(__QDISC_STATE_DEACTIVATED,
495 &qdisc_root_sleeping(wd->qdisc)->state))
498 wd->qdisc->flags |= TCQ_F_THROTTLED;
499 time = ktime_set(0, 0);
500 time = ktime_add_ns(time, PSCHED_TICKS2NS(expires));
501 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
503 EXPORT_SYMBOL(qdisc_watchdog_schedule);
505 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
507 hrtimer_cancel(&wd->timer);
508 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
510 EXPORT_SYMBOL(qdisc_watchdog_cancel);
512 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
514 unsigned int size = n * sizeof(struct hlist_head), i;
515 struct hlist_head *h;
517 if (size <= PAGE_SIZE)
518 h = kmalloc(size, GFP_KERNEL);
520 h = (struct hlist_head *)
521 __get_free_pages(GFP_KERNEL, get_order(size));
524 for (i = 0; i < n; i++)
525 INIT_HLIST_HEAD(&h[i]);
530 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
532 unsigned int size = n * sizeof(struct hlist_head);
534 if (size <= PAGE_SIZE)
537 free_pages((unsigned long)h, get_order(size));
540 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
542 struct Qdisc_class_common *cl;
543 struct hlist_node *n, *next;
544 struct hlist_head *nhash, *ohash;
545 unsigned int nsize, nmask, osize;
548 /* Rehash when load factor exceeds 0.75 */
549 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
551 nsize = clhash->hashsize * 2;
553 nhash = qdisc_class_hash_alloc(nsize);
557 ohash = clhash->hash;
558 osize = clhash->hashsize;
561 for (i = 0; i < osize; i++) {
562 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
563 h = qdisc_class_hash(cl->classid, nmask);
564 hlist_add_head(&cl->hnode, &nhash[h]);
567 clhash->hash = nhash;
568 clhash->hashsize = nsize;
569 clhash->hashmask = nmask;
570 sch_tree_unlock(sch);
572 qdisc_class_hash_free(ohash, osize);
574 EXPORT_SYMBOL(qdisc_class_hash_grow);
576 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
578 unsigned int size = 4;
580 clhash->hash = qdisc_class_hash_alloc(size);
581 if (clhash->hash == NULL)
583 clhash->hashsize = size;
584 clhash->hashmask = size - 1;
585 clhash->hashelems = 0;
588 EXPORT_SYMBOL(qdisc_class_hash_init);
590 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
592 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
594 EXPORT_SYMBOL(qdisc_class_hash_destroy);
596 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
597 struct Qdisc_class_common *cl)
601 INIT_HLIST_NODE(&cl->hnode);
602 h = qdisc_class_hash(cl->classid, clhash->hashmask);
603 hlist_add_head(&cl->hnode, &clhash->hash[h]);
606 EXPORT_SYMBOL(qdisc_class_hash_insert);
608 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
609 struct Qdisc_class_common *cl)
611 hlist_del(&cl->hnode);
614 EXPORT_SYMBOL(qdisc_class_hash_remove);
616 /* Allocate an unique handle from space managed by kernel */
618 static u32 qdisc_alloc_handle(struct net_device *dev)
621 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
624 autohandle += TC_H_MAKE(0x10000U, 0);
625 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
626 autohandle = TC_H_MAKE(0x80000000U, 0);
627 } while (qdisc_lookup(dev, autohandle) && --i > 0);
629 return i > 0 ? autohandle : 0;
632 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
634 const struct Qdisc_class_ops *cops;
640 while ((parentid = sch->parent)) {
641 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
644 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
646 WARN_ON(parentid != TC_H_ROOT);
649 cops = sch->ops->cl_ops;
650 if (cops->qlen_notify) {
651 cl = cops->get(sch, parentid);
652 cops->qlen_notify(sch, cl);
658 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
660 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
661 struct nlmsghdr *n, u32 clid,
662 struct Qdisc *old, struct Qdisc *new)
665 qdisc_notify(net, skb, n, clid, old, new);
671 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
674 * When appropriate send a netlink notification using 'skb'
677 * On success, destroy old qdisc.
680 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
681 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
682 struct Qdisc *new, struct Qdisc *old)
684 struct Qdisc *q = old;
685 struct net *net = dev_net(dev);
688 if (parent == NULL) {
689 unsigned int i, num_q, ingress;
692 num_q = dev->num_tx_queues;
693 if ((q && q->flags & TCQ_F_INGRESS) ||
694 (new && new->flags & TCQ_F_INGRESS)) {
697 if (!dev_ingress_queue(dev))
701 if (dev->flags & IFF_UP)
704 if (new && new->ops->attach) {
705 new->ops->attach(new);
709 for (i = 0; i < num_q; i++) {
710 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
713 dev_queue = netdev_get_tx_queue(dev, i);
715 old = dev_graft_qdisc(dev_queue, new);
717 atomic_inc(&new->refcnt);
724 notify_and_destroy(net, skb, n, classid,
726 if (new && !new->ops->attach)
727 atomic_inc(&new->refcnt);
728 dev->qdisc = new ? : &noop_qdisc;
730 notify_and_destroy(net, skb, n, classid, old, new);
733 if (dev->flags & IFF_UP)
736 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
739 if (cops && cops->graft) {
740 unsigned long cl = cops->get(parent, classid);
742 err = cops->graft(parent, cl, new, &old);
743 cops->put(parent, cl);
748 notify_and_destroy(net, skb, n, classid, old, new);
753 /* lockdep annotation is needed for ingress; egress gets it only for name */
754 static struct lock_class_key qdisc_tx_lock;
755 static struct lock_class_key qdisc_rx_lock;
758 Allocate and initialize new qdisc.
760 Parameters are passed via opt.
763 static struct Qdisc *
764 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
765 struct Qdisc *p, u32 parent, u32 handle,
766 struct nlattr **tca, int *errp)
769 struct nlattr *kind = tca[TCA_KIND];
771 struct Qdisc_ops *ops;
772 struct qdisc_size_table *stab;
774 ops = qdisc_lookup_ops(kind);
775 #ifdef CONFIG_MODULES
776 if (ops == NULL && kind != NULL) {
778 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
779 /* We dropped the RTNL semaphore in order to
780 * perform the module load. So, even if we
781 * succeeded in loading the module we have to
782 * tell the caller to replay the request. We
783 * indicate this using -EAGAIN.
784 * We replay the request because the device may
785 * go away in the mean time.
788 request_module("sch_%s", name);
790 ops = qdisc_lookup_ops(kind);
792 /* We will try again qdisc_lookup_ops,
793 * so don't keep a reference.
795 module_put(ops->owner);
807 sch = qdisc_alloc(dev_queue, ops);
813 sch->parent = parent;
815 if (handle == TC_H_INGRESS) {
816 sch->flags |= TCQ_F_INGRESS;
817 handle = TC_H_MAKE(TC_H_INGRESS, 0);
818 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
821 handle = qdisc_alloc_handle(dev);
826 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
829 sch->handle = handle;
831 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
833 stab = qdisc_get_stab(tca[TCA_STAB]);
841 spinlock_t *root_lock;
844 if (sch->flags & TCQ_F_MQROOT)
847 if ((sch->parent != TC_H_ROOT) &&
848 !(sch->flags & TCQ_F_INGRESS) &&
849 (!p || !(p->flags & TCQ_F_MQROOT)))
850 root_lock = qdisc_root_sleeping_lock(sch);
852 root_lock = qdisc_lock(sch);
854 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
855 root_lock, tca[TCA_RATE]);
866 kfree((char *) sch - sch->padded);
868 module_put(ops->owner);
875 * Any broken qdiscs that would require a ops->reset() here?
876 * The qdisc was never in action so it shouldn't be necessary.
878 qdisc_put_stab(sch->stab);
884 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
886 struct qdisc_size_table *stab = NULL;
889 if (tca[TCA_OPTIONS]) {
890 if (sch->ops->change == NULL)
892 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
898 stab = qdisc_get_stab(tca[TCA_STAB]);
900 return PTR_ERR(stab);
903 qdisc_put_stab(sch->stab);
907 /* NB: ignores errors from replace_estimator
908 because change can't be undone. */
909 if (sch->flags & TCQ_F_MQROOT)
911 gen_replace_estimator(&sch->bstats, &sch->rate_est,
912 qdisc_root_sleeping_lock(sch),
919 struct check_loop_arg {
920 struct qdisc_walker w;
925 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
927 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
929 struct check_loop_arg arg;
931 if (q->ops->cl_ops == NULL)
934 arg.w.stop = arg.w.skip = arg.w.count = 0;
935 arg.w.fn = check_loop_fn;
938 q->ops->cl_ops->walk(q, &arg.w);
939 return arg.w.stop ? -ELOOP : 0;
943 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
946 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
947 struct check_loop_arg *arg = (struct check_loop_arg *)w;
949 leaf = cops->leaf(q, cl);
951 if (leaf == arg->p || arg->depth > 7)
953 return check_loop(leaf, arg->p, arg->depth + 1);
962 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
964 struct net *net = sock_net(skb->sk);
965 struct tcmsg *tcm = NLMSG_DATA(n);
966 struct nlattr *tca[TCA_MAX + 1];
967 struct net_device *dev;
968 u32 clid = tcm->tcm_parent;
969 struct Qdisc *q = NULL;
970 struct Qdisc *p = NULL;
973 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
977 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
982 if (clid != TC_H_ROOT) {
983 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
984 p = qdisc_lookup(dev, TC_H_MAJ(clid));
987 q = qdisc_leaf(p, clid);
988 } else if (dev_ingress_queue(dev)) {
989 q = dev_ingress_queue(dev)->qdisc_sleeping;
997 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1000 q = qdisc_lookup(dev, tcm->tcm_handle);
1005 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1008 if (n->nlmsg_type == RTM_DELQDISC) {
1013 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1017 qdisc_notify(net, skb, n, clid, NULL, q);
1023 * Create/change qdisc.
1026 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1028 struct net *net = sock_net(skb->sk);
1030 struct nlattr *tca[TCA_MAX + 1];
1031 struct net_device *dev;
1033 struct Qdisc *q, *p;
1037 /* Reinit, just in case something touches this. */
1038 tcm = NLMSG_DATA(n);
1039 clid = tcm->tcm_parent;
1042 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1046 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1051 if (clid != TC_H_ROOT) {
1052 if (clid != TC_H_INGRESS) {
1053 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1056 q = qdisc_leaf(p, clid);
1057 } else if (dev_ingress_queue_create(dev)) {
1058 q = dev_ingress_queue(dev)->qdisc_sleeping;
1064 /* It may be default qdisc, ignore it */
1065 if (q && q->handle == 0)
1068 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1069 if (tcm->tcm_handle) {
1070 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1072 if (TC_H_MIN(tcm->tcm_handle))
1074 q = qdisc_lookup(dev, tcm->tcm_handle);
1076 goto create_n_graft;
1077 if (n->nlmsg_flags & NLM_F_EXCL)
1079 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1082 (p && check_loop(q, p, 0)))
1084 atomic_inc(&q->refcnt);
1088 goto create_n_graft;
1090 /* This magic test requires explanation.
1092 * We know, that some child q is already
1093 * attached to this parent and have choice:
1094 * either to change it or to create/graft new one.
1096 * 1. We are allowed to create/graft only
1097 * if CREATE and REPLACE flags are set.
1099 * 2. If EXCL is set, requestor wanted to say,
1100 * that qdisc tcm_handle is not expected
1101 * to exist, so that we choose create/graft too.
1103 * 3. The last case is when no flags are set.
1104 * Alas, it is sort of hole in API, we
1105 * cannot decide what to do unambiguously.
1106 * For now we select create/graft, if
1107 * user gave KIND, which does not match existing.
1109 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1110 (n->nlmsg_flags & NLM_F_REPLACE) &&
1111 ((n->nlmsg_flags & NLM_F_EXCL) ||
1113 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1114 goto create_n_graft;
1118 if (!tcm->tcm_handle)
1120 q = qdisc_lookup(dev, tcm->tcm_handle);
1123 /* Change qdisc parameters */
1126 if (n->nlmsg_flags & NLM_F_EXCL)
1128 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1130 err = qdisc_change(q, tca);
1132 qdisc_notify(net, skb, n, clid, NULL, q);
1136 if (!(n->nlmsg_flags & NLM_F_CREATE))
1138 if (clid == TC_H_INGRESS) {
1139 if (dev_ingress_queue(dev))
1140 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1141 tcm->tcm_parent, tcm->tcm_parent,
1146 struct netdev_queue *dev_queue;
1148 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1149 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1151 dev_queue = p->dev_queue;
1153 dev_queue = netdev_get_tx_queue(dev, 0);
1155 q = qdisc_create(dev, dev_queue, p,
1156 tcm->tcm_parent, tcm->tcm_handle,
1166 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1176 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1177 u32 pid, u32 seq, u16 flags, int event)
1180 struct nlmsghdr *nlh;
1181 unsigned char *b = skb_tail_pointer(skb);
1184 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1185 tcm = NLMSG_DATA(nlh);
1186 tcm->tcm_family = AF_UNSPEC;
1189 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1190 tcm->tcm_parent = clid;
1191 tcm->tcm_handle = q->handle;
1192 tcm->tcm_info = atomic_read(&q->refcnt);
1193 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1194 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1195 goto nla_put_failure;
1196 q->qstats.qlen = q->q.qlen;
1198 if (q->stab && qdisc_dump_stab(skb, q->stab) < 0)
1199 goto nla_put_failure;
1201 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1202 qdisc_root_sleeping_lock(q), &d) < 0)
1203 goto nla_put_failure;
1205 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1206 goto nla_put_failure;
1208 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1209 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1210 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1211 goto nla_put_failure;
1213 if (gnet_stats_finish_copy(&d) < 0)
1214 goto nla_put_failure;
1216 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1225 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1227 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1230 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1231 struct nlmsghdr *n, u32 clid,
1232 struct Qdisc *old, struct Qdisc *new)
1234 struct sk_buff *skb;
1235 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1237 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1241 if (old && !tc_qdisc_dump_ignore(old)) {
1242 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq,
1243 0, RTM_DELQDISC) < 0)
1246 if (new && !tc_qdisc_dump_ignore(new)) {
1247 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq,
1248 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1253 return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
1254 n->nlmsg_flags & NLM_F_ECHO);
1261 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1262 struct netlink_callback *cb,
1263 int *q_idx_p, int s_q_idx)
1265 int ret = 0, q_idx = *q_idx_p;
1272 if (q_idx < s_q_idx) {
1275 if (!tc_qdisc_dump_ignore(q) &&
1276 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1277 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1281 list_for_each_entry(q, &root->list, list) {
1282 if (q_idx < s_q_idx) {
1286 if (!tc_qdisc_dump_ignore(q) &&
1287 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1288 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1301 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1303 struct net *net = sock_net(skb->sk);
1306 struct net_device *dev;
1308 s_idx = cb->args[0];
1309 s_q_idx = q_idx = cb->args[1];
1313 for_each_netdev_rcu(net, dev) {
1314 struct netdev_queue *dev_queue;
1322 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1325 dev_queue = dev_ingress_queue(dev);
1327 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1328 &q_idx, s_q_idx) < 0)
1339 cb->args[1] = q_idx;
1346 /************************************************
1347 * Traffic classes manipulation. *
1348 ************************************************/
1352 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1354 struct net *net = sock_net(skb->sk);
1355 struct tcmsg *tcm = NLMSG_DATA(n);
1356 struct nlattr *tca[TCA_MAX + 1];
1357 struct net_device *dev;
1358 struct Qdisc *q = NULL;
1359 const struct Qdisc_class_ops *cops;
1360 unsigned long cl = 0;
1361 unsigned long new_cl;
1362 u32 pid = tcm->tcm_parent;
1363 u32 clid = tcm->tcm_handle;
1364 u32 qid = TC_H_MAJ(clid);
1367 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1371 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1376 parent == TC_H_UNSPEC - unspecified parent.
1377 parent == TC_H_ROOT - class is root, which has no parent.
1378 parent == X:0 - parent is root class.
1379 parent == X:Y - parent is a node in hierarchy.
1380 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1382 handle == 0:0 - generate handle from kernel pool.
1383 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1384 handle == X:Y - clear.
1385 handle == X:0 - root class.
1388 /* Step 1. Determine qdisc handle X:0 */
1390 if (pid != TC_H_ROOT) {
1391 u32 qid1 = TC_H_MAJ(pid);
1394 /* If both majors are known, they must be identical. */
1399 } else if (qid == 0)
1400 qid = dev->qdisc->handle;
1402 /* Now qid is genuine qdisc handle consistent
1403 * both with parent and child.
1405 * TC_H_MAJ(pid) still may be unspecified, complete it now.
1408 pid = TC_H_MAKE(qid, pid);
1411 qid = dev->qdisc->handle;
1414 /* OK. Locate qdisc */
1415 q = qdisc_lookup(dev, qid);
1419 /* An check that it supports classes */
1420 cops = q->ops->cl_ops;
1424 /* Now try to get class */
1426 if (pid == TC_H_ROOT)
1429 clid = TC_H_MAKE(qid, clid);
1432 cl = cops->get(q, clid);
1436 if (n->nlmsg_type != RTM_NEWTCLASS ||
1437 !(n->nlmsg_flags & NLM_F_CREATE))
1440 switch (n->nlmsg_type) {
1443 if (n->nlmsg_flags & NLM_F_EXCL)
1449 err = cops->delete(q, cl);
1451 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1454 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1465 err = cops->change(q, clid, pid, tca, &new_cl);
1467 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1477 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1479 u32 pid, u32 seq, u16 flags, int event)
1482 struct nlmsghdr *nlh;
1483 unsigned char *b = skb_tail_pointer(skb);
1485 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1487 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1488 tcm = NLMSG_DATA(nlh);
1489 tcm->tcm_family = AF_UNSPEC;
1492 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1493 tcm->tcm_parent = q->handle;
1494 tcm->tcm_handle = q->handle;
1496 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1497 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1498 goto nla_put_failure;
1500 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1501 qdisc_root_sleeping_lock(q), &d) < 0)
1502 goto nla_put_failure;
1504 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1505 goto nla_put_failure;
1507 if (gnet_stats_finish_copy(&d) < 0)
1508 goto nla_put_failure;
1510 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1519 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1520 struct nlmsghdr *n, struct Qdisc *q,
1521 unsigned long cl, int event)
1523 struct sk_buff *skb;
1524 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1526 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1530 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1535 return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
1536 n->nlmsg_flags & NLM_F_ECHO);
1539 struct qdisc_dump_args {
1540 struct qdisc_walker w;
1541 struct sk_buff *skb;
1542 struct netlink_callback *cb;
1545 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1547 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1549 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1550 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1553 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1554 struct tcmsg *tcm, struct netlink_callback *cb,
1557 struct qdisc_dump_args arg;
1559 if (tc_qdisc_dump_ignore(q) ||
1560 *t_p < s_t || !q->ops->cl_ops ||
1562 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1567 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1568 arg.w.fn = qdisc_class_dump;
1572 arg.w.skip = cb->args[1];
1574 q->ops->cl_ops->walk(q, &arg.w);
1575 cb->args[1] = arg.w.count;
1582 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1583 struct tcmsg *tcm, struct netlink_callback *cb,
1591 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1594 list_for_each_entry(q, &root->list, list) {
1595 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1602 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1604 struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
1605 struct net *net = sock_net(skb->sk);
1606 struct netdev_queue *dev_queue;
1607 struct net_device *dev;
1610 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1612 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1619 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1622 dev_queue = dev_ingress_queue(dev);
1624 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1635 /* Main classifier routine: scans classifier chain attached
1636 * to this qdisc, (optionally) tests for protocol and asks
1637 * specific classifiers.
1639 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1640 struct tcf_result *res)
1642 __be16 protocol = skb->protocol;
1645 for (; tp; tp = tp->next) {
1646 if (tp->protocol != protocol &&
1647 tp->protocol != htons(ETH_P_ALL))
1649 err = tp->classify(skb, tp, res);
1652 #ifdef CONFIG_NET_CLS_ACT
1653 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1654 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1661 EXPORT_SYMBOL(tc_classify_compat);
1663 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1664 struct tcf_result *res)
1668 #ifdef CONFIG_NET_CLS_ACT
1669 struct tcf_proto *otp = tp;
1672 protocol = skb->protocol;
1674 err = tc_classify_compat(skb, tp, res);
1675 #ifdef CONFIG_NET_CLS_ACT
1676 if (err == TC_ACT_RECLASSIFY) {
1677 u32 verd = G_TC_VERD(skb->tc_verd);
1680 if (verd++ >= MAX_REC_LOOP) {
1681 if (net_ratelimit())
1682 pr_notice("%s: packet reclassify loop"
1683 " rule prio %u protocol %02x\n",
1686 ntohs(tp->protocol));
1689 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1695 EXPORT_SYMBOL(tc_classify);
1697 void tcf_destroy(struct tcf_proto *tp)
1699 tp->ops->destroy(tp);
1700 module_put(tp->ops->owner);
1704 void tcf_destroy_chain(struct tcf_proto **fl)
1706 struct tcf_proto *tp;
1708 while ((tp = *fl) != NULL) {
1713 EXPORT_SYMBOL(tcf_destroy_chain);
1715 #ifdef CONFIG_PROC_FS
1716 static int psched_show(struct seq_file *seq, void *v)
1720 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1721 seq_printf(seq, "%08x %08x %08x %08x\n",
1722 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1724 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1729 static int psched_open(struct inode *inode, struct file *file)
1731 return single_open(file, psched_show, NULL);
1734 static const struct file_operations psched_fops = {
1735 .owner = THIS_MODULE,
1736 .open = psched_open,
1738 .llseek = seq_lseek,
1739 .release = single_release,
1742 static int __net_init psched_net_init(struct net *net)
1744 struct proc_dir_entry *e;
1746 e = proc_net_fops_create(net, "psched", 0, &psched_fops);
1753 static void __net_exit psched_net_exit(struct net *net)
1755 proc_net_remove(net, "psched");
1758 static int __net_init psched_net_init(struct net *net)
1763 static void __net_exit psched_net_exit(struct net *net)
1768 static struct pernet_operations psched_net_ops = {
1769 .init = psched_net_init,
1770 .exit = psched_net_exit,
1773 static int __init pktsched_init(void)
1777 err = register_pernet_subsys(&psched_net_ops);
1779 pr_err("pktsched_init: "
1780 "cannot initialize per netns operations\n");
1784 register_qdisc(&pfifo_qdisc_ops);
1785 register_qdisc(&bfifo_qdisc_ops);
1786 register_qdisc(&pfifo_head_drop_qdisc_ops);
1787 register_qdisc(&mq_qdisc_ops);
1789 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1790 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1791 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1792 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1793 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1794 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1799 subsys_initcall(pktsched_init);