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 like dequeue but without removing a packet from the queue
106 returns qdisc to initial state: purge all buffers, clear all
107 timers, counters (except for statistics) etc.
111 initializes newly created qdisc.
115 destroys resources allocated by init and during lifetime of qdisc.
119 changes qdisc parameters.
122 /* Protects list of registered TC modules. It is pure SMP lock. */
123 static DEFINE_RWLOCK(qdisc_mod_lock);
126 /************************************************
127 * Queueing disciplines manipulation. *
128 ************************************************/
131 /* The list of all installed queueing disciplines. */
133 static struct Qdisc_ops *qdisc_base;
135 /* Register/uregister queueing discipline */
137 int register_qdisc(struct Qdisc_ops *qops)
139 struct Qdisc_ops *q, **qp;
142 write_lock(&qdisc_mod_lock);
143 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
144 if (!strcmp(qops->id, q->id))
147 if (qops->enqueue == NULL)
148 qops->enqueue = noop_qdisc_ops.enqueue;
149 if (qops->peek == NULL) {
150 if (qops->dequeue == NULL) {
151 qops->peek = noop_qdisc_ops.peek;
157 if (qops->dequeue == NULL)
158 qops->dequeue = noop_qdisc_ops.dequeue;
164 write_unlock(&qdisc_mod_lock);
167 EXPORT_SYMBOL(register_qdisc);
169 int unregister_qdisc(struct Qdisc_ops *qops)
171 struct Qdisc_ops *q, **qp;
174 write_lock(&qdisc_mod_lock);
175 for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next)
183 write_unlock(&qdisc_mod_lock);
186 EXPORT_SYMBOL(unregister_qdisc);
188 /* We know handle. Find qdisc among all qdisc's attached to device
189 (root qdisc, all its children, children of children etc.)
192 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
196 if (!(root->flags & TCQ_F_BUILTIN) &&
197 root->handle == handle)
200 list_for_each_entry(q, &root->list, list) {
201 if (q->handle == handle)
207 static void qdisc_list_add(struct Qdisc *q)
209 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
210 list_add_tail(&q->list, &qdisc_root_sleeping(q)->list);
213 void qdisc_list_del(struct Qdisc *q)
215 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
218 EXPORT_SYMBOL(qdisc_list_del);
220 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
225 for (i = 0; i < dev->num_tx_queues; i++) {
226 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
227 struct Qdisc *txq_root = txq->qdisc_sleeping;
229 q = qdisc_match_from_root(txq_root, handle);
234 q = qdisc_match_from_root(dev->rx_queue.qdisc_sleeping, handle);
239 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
243 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
247 cl = cops->get(p, classid);
251 leaf = cops->leaf(p, cl);
256 /* Find queueing discipline by name */
258 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
260 struct Qdisc_ops *q = NULL;
263 read_lock(&qdisc_mod_lock);
264 for (q = qdisc_base; q; q = q->next) {
265 if (nla_strcmp(kind, q->id) == 0) {
266 if (!try_module_get(q->owner))
271 read_unlock(&qdisc_mod_lock);
276 static struct qdisc_rate_table *qdisc_rtab_list;
278 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
280 struct qdisc_rate_table *rtab;
282 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
283 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
289 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
290 nla_len(tab) != TC_RTAB_SIZE)
293 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
297 memcpy(rtab->data, nla_data(tab), 1024);
298 rtab->next = qdisc_rtab_list;
299 qdisc_rtab_list = rtab;
303 EXPORT_SYMBOL(qdisc_get_rtab);
305 void qdisc_put_rtab(struct qdisc_rate_table *tab)
307 struct qdisc_rate_table *rtab, **rtabp;
309 if (!tab || --tab->refcnt)
312 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
320 EXPORT_SYMBOL(qdisc_put_rtab);
322 static LIST_HEAD(qdisc_stab_list);
323 static DEFINE_SPINLOCK(qdisc_stab_lock);
325 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
326 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
327 [TCA_STAB_DATA] = { .type = NLA_BINARY },
330 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
332 struct nlattr *tb[TCA_STAB_MAX + 1];
333 struct qdisc_size_table *stab;
334 struct tc_sizespec *s;
335 unsigned int tsize = 0;
339 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
342 if (!tb[TCA_STAB_BASE])
343 return ERR_PTR(-EINVAL);
345 s = nla_data(tb[TCA_STAB_BASE]);
348 if (!tb[TCA_STAB_DATA])
349 return ERR_PTR(-EINVAL);
350 tab = nla_data(tb[TCA_STAB_DATA]);
351 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
354 if (!s || tsize != s->tsize || (!tab && tsize > 0))
355 return ERR_PTR(-EINVAL);
357 spin_lock(&qdisc_stab_lock);
359 list_for_each_entry(stab, &qdisc_stab_list, list) {
360 if (memcmp(&stab->szopts, s, sizeof(*s)))
362 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
365 spin_unlock(&qdisc_stab_lock);
369 spin_unlock(&qdisc_stab_lock);
371 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
373 return ERR_PTR(-ENOMEM);
378 memcpy(stab->data, tab, tsize * sizeof(u16));
380 spin_lock(&qdisc_stab_lock);
381 list_add_tail(&stab->list, &qdisc_stab_list);
382 spin_unlock(&qdisc_stab_lock);
387 void qdisc_put_stab(struct qdisc_size_table *tab)
392 spin_lock(&qdisc_stab_lock);
394 if (--tab->refcnt == 0) {
395 list_del(&tab->list);
399 spin_unlock(&qdisc_stab_lock);
401 EXPORT_SYMBOL(qdisc_put_stab);
403 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
407 nest = nla_nest_start(skb, TCA_STAB);
409 goto nla_put_failure;
410 NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts);
411 nla_nest_end(skb, nest);
419 void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab)
423 pkt_len = skb->len + stab->szopts.overhead;
424 if (unlikely(!stab->szopts.tsize))
427 slot = pkt_len + stab->szopts.cell_align;
428 if (unlikely(slot < 0))
431 slot >>= stab->szopts.cell_log;
432 if (likely(slot < stab->szopts.tsize))
433 pkt_len = stab->data[slot];
435 pkt_len = stab->data[stab->szopts.tsize - 1] *
436 (slot / stab->szopts.tsize) +
437 stab->data[slot % stab->szopts.tsize];
439 pkt_len <<= stab->szopts.size_log;
441 if (unlikely(pkt_len < 1))
443 qdisc_skb_cb(skb)->pkt_len = pkt_len;
445 EXPORT_SYMBOL(qdisc_calculate_pkt_len);
447 void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
449 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
451 "%s: %s qdisc %X: is non-work-conserving?\n",
452 txt, qdisc->ops->id, qdisc->handle >> 16);
453 qdisc->flags |= TCQ_F_WARN_NONWC;
456 EXPORT_SYMBOL(qdisc_warn_nonwc);
458 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
460 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
463 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
464 __netif_schedule(qdisc_root(wd->qdisc));
466 return HRTIMER_NORESTART;
469 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
471 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
472 wd->timer.function = qdisc_watchdog;
475 EXPORT_SYMBOL(qdisc_watchdog_init);
477 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
481 if (test_bit(__QDISC_STATE_DEACTIVATED,
482 &qdisc_root_sleeping(wd->qdisc)->state))
485 wd->qdisc->flags |= TCQ_F_THROTTLED;
486 time = ktime_set(0, 0);
487 time = ktime_add_ns(time, PSCHED_US2NS(expires));
488 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
490 EXPORT_SYMBOL(qdisc_watchdog_schedule);
492 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
494 hrtimer_cancel(&wd->timer);
495 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
497 EXPORT_SYMBOL(qdisc_watchdog_cancel);
499 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
501 unsigned int size = n * sizeof(struct hlist_head), i;
502 struct hlist_head *h;
504 if (size <= PAGE_SIZE)
505 h = kmalloc(size, GFP_KERNEL);
507 h = (struct hlist_head *)
508 __get_free_pages(GFP_KERNEL, get_order(size));
511 for (i = 0; i < n; i++)
512 INIT_HLIST_HEAD(&h[i]);
517 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
519 unsigned int size = n * sizeof(struct hlist_head);
521 if (size <= PAGE_SIZE)
524 free_pages((unsigned long)h, get_order(size));
527 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
529 struct Qdisc_class_common *cl;
530 struct hlist_node *n, *next;
531 struct hlist_head *nhash, *ohash;
532 unsigned int nsize, nmask, osize;
535 /* Rehash when load factor exceeds 0.75 */
536 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
538 nsize = clhash->hashsize * 2;
540 nhash = qdisc_class_hash_alloc(nsize);
544 ohash = clhash->hash;
545 osize = clhash->hashsize;
548 for (i = 0; i < osize; i++) {
549 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
550 h = qdisc_class_hash(cl->classid, nmask);
551 hlist_add_head(&cl->hnode, &nhash[h]);
554 clhash->hash = nhash;
555 clhash->hashsize = nsize;
556 clhash->hashmask = nmask;
557 sch_tree_unlock(sch);
559 qdisc_class_hash_free(ohash, osize);
561 EXPORT_SYMBOL(qdisc_class_hash_grow);
563 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
565 unsigned int size = 4;
567 clhash->hash = qdisc_class_hash_alloc(size);
568 if (clhash->hash == NULL)
570 clhash->hashsize = size;
571 clhash->hashmask = size - 1;
572 clhash->hashelems = 0;
575 EXPORT_SYMBOL(qdisc_class_hash_init);
577 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
579 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
581 EXPORT_SYMBOL(qdisc_class_hash_destroy);
583 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
584 struct Qdisc_class_common *cl)
588 INIT_HLIST_NODE(&cl->hnode);
589 h = qdisc_class_hash(cl->classid, clhash->hashmask);
590 hlist_add_head(&cl->hnode, &clhash->hash[h]);
593 EXPORT_SYMBOL(qdisc_class_hash_insert);
595 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
596 struct Qdisc_class_common *cl)
598 hlist_del(&cl->hnode);
601 EXPORT_SYMBOL(qdisc_class_hash_remove);
603 /* Allocate an unique handle from space managed by kernel */
605 static u32 qdisc_alloc_handle(struct net_device *dev)
608 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
611 autohandle += TC_H_MAKE(0x10000U, 0);
612 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
613 autohandle = TC_H_MAKE(0x80000000U, 0);
614 } while (qdisc_lookup(dev, autohandle) && --i > 0);
616 return i>0 ? autohandle : 0;
619 /* Attach toplevel qdisc to device queue. */
621 static struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
624 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
625 spinlock_t *root_lock;
627 root_lock = qdisc_lock(oqdisc);
628 spin_lock_bh(root_lock);
630 /* Prune old scheduler */
631 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
634 /* ... and graft new one */
637 dev_queue->qdisc_sleeping = qdisc;
638 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
640 spin_unlock_bh(root_lock);
645 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
647 const struct Qdisc_class_ops *cops;
653 while ((parentid = sch->parent)) {
654 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
657 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
659 WARN_ON(parentid != TC_H_ROOT);
662 cops = sch->ops->cl_ops;
663 if (cops->qlen_notify) {
664 cl = cops->get(sch, parentid);
665 cops->qlen_notify(sch, cl);
671 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
673 static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid,
674 struct Qdisc *old, struct Qdisc *new)
677 qdisc_notify(skb, n, clid, old, new);
683 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
686 * When appropriate send a netlink notification using 'skb'
689 * On success, destroy old qdisc.
692 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
693 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
694 struct Qdisc *new, struct Qdisc *old)
696 struct Qdisc *q = old;
699 if (parent == NULL) {
700 unsigned int i, num_q, ingress;
703 num_q = dev->num_tx_queues;
704 if ((q && q->flags & TCQ_F_INGRESS) ||
705 (new && new->flags & TCQ_F_INGRESS)) {
710 if (dev->flags & IFF_UP)
713 for (i = 0; i < num_q; i++) {
714 struct netdev_queue *dev_queue = &dev->rx_queue;
717 dev_queue = netdev_get_tx_queue(dev, i);
719 old = dev_graft_qdisc(dev_queue, new);
721 atomic_inc(&new->refcnt);
723 notify_and_destroy(skb, n, classid, old, new);
726 if (dev->flags & IFF_UP)
729 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
734 unsigned long cl = cops->get(parent, classid);
736 err = cops->graft(parent, cl, new, &old);
737 cops->put(parent, cl);
741 notify_and_destroy(skb, n, classid, old, new);
746 /* lockdep annotation is needed for ingress; egress gets it only for name */
747 static struct lock_class_key qdisc_tx_lock;
748 static struct lock_class_key qdisc_rx_lock;
751 Allocate and initialize new qdisc.
753 Parameters are passed via opt.
756 static struct Qdisc *
757 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
758 u32 parent, u32 handle, struct nlattr **tca, int *errp)
761 struct nlattr *kind = tca[TCA_KIND];
763 struct Qdisc_ops *ops;
764 struct qdisc_size_table *stab;
766 ops = qdisc_lookup_ops(kind);
767 #ifdef CONFIG_MODULES
768 if (ops == NULL && kind != NULL) {
770 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
771 /* We dropped the RTNL semaphore in order to
772 * perform the module load. So, even if we
773 * succeeded in loading the module we have to
774 * tell the caller to replay the request. We
775 * indicate this using -EAGAIN.
776 * We replay the request because the device may
777 * go away in the mean time.
780 request_module("sch_%s", name);
782 ops = qdisc_lookup_ops(kind);
784 /* We will try again qdisc_lookup_ops,
785 * so don't keep a reference.
787 module_put(ops->owner);
799 sch = qdisc_alloc(dev_queue, ops);
805 sch->parent = parent;
807 if (handle == TC_H_INGRESS) {
808 sch->flags |= TCQ_F_INGRESS;
809 handle = TC_H_MAKE(TC_H_INGRESS, 0);
810 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
813 handle = qdisc_alloc_handle(dev);
818 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
821 sch->handle = handle;
823 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
825 stab = qdisc_get_stab(tca[TCA_STAB]);
833 spinlock_t *root_lock;
835 if ((sch->parent != TC_H_ROOT) &&
836 !(sch->flags & TCQ_F_INGRESS))
837 root_lock = qdisc_root_sleeping_lock(sch);
839 root_lock = qdisc_lock(sch);
841 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
842 root_lock, tca[TCA_RATE]);
845 * Any broken qdiscs that would require
846 * a ops->reset() here? The qdisc was never
847 * in action so it shouldn't be necessary.
860 qdisc_put_stab(sch->stab);
862 kfree((char *) sch - sch->padded);
864 module_put(ops->owner);
870 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
872 struct qdisc_size_table *stab = NULL;
875 if (tca[TCA_OPTIONS]) {
876 if (sch->ops->change == NULL)
878 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
884 stab = qdisc_get_stab(tca[TCA_STAB]);
886 return PTR_ERR(stab);
889 qdisc_put_stab(sch->stab);
893 /* NB: ignores errors from replace_estimator
894 because change can't be undone. */
895 gen_replace_estimator(&sch->bstats, &sch->rate_est,
896 qdisc_root_sleeping_lock(sch),
902 struct check_loop_arg
904 struct qdisc_walker w;
909 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
911 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
913 struct check_loop_arg arg;
915 if (q->ops->cl_ops == NULL)
918 arg.w.stop = arg.w.skip = arg.w.count = 0;
919 arg.w.fn = check_loop_fn;
922 q->ops->cl_ops->walk(q, &arg.w);
923 return arg.w.stop ? -ELOOP : 0;
927 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
930 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
931 struct check_loop_arg *arg = (struct check_loop_arg *)w;
933 leaf = cops->leaf(q, cl);
935 if (leaf == arg->p || arg->depth > 7)
937 return check_loop(leaf, arg->p, arg->depth + 1);
946 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
948 struct net *net = sock_net(skb->sk);
949 struct tcmsg *tcm = NLMSG_DATA(n);
950 struct nlattr *tca[TCA_MAX + 1];
951 struct net_device *dev;
952 u32 clid = tcm->tcm_parent;
953 struct Qdisc *q = NULL;
954 struct Qdisc *p = NULL;
957 if (net != &init_net)
960 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
963 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
968 if (clid != TC_H_ROOT) {
969 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
970 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
972 q = qdisc_leaf(p, clid);
973 } else { /* ingress */
974 q = dev->rx_queue.qdisc_sleeping;
977 struct netdev_queue *dev_queue;
978 dev_queue = netdev_get_tx_queue(dev, 0);
979 q = dev_queue->qdisc_sleeping;
984 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
987 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
991 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
994 if (n->nlmsg_type == RTM_DELQDISC) {
999 if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0)
1002 qdisc_notify(skb, n, clid, NULL, q);
1008 Create/change qdisc.
1011 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1013 struct net *net = sock_net(skb->sk);
1015 struct nlattr *tca[TCA_MAX + 1];
1016 struct net_device *dev;
1018 struct Qdisc *q, *p;
1021 if (net != &init_net)
1025 /* Reinit, just in case something touches this. */
1026 tcm = NLMSG_DATA(n);
1027 clid = tcm->tcm_parent;
1030 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1033 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1038 if (clid != TC_H_ROOT) {
1039 if (clid != TC_H_INGRESS) {
1040 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
1042 q = qdisc_leaf(p, clid);
1043 } else { /*ingress */
1044 q = dev->rx_queue.qdisc_sleeping;
1047 struct netdev_queue *dev_queue;
1048 dev_queue = netdev_get_tx_queue(dev, 0);
1049 q = dev_queue->qdisc_sleeping;
1052 /* It may be default qdisc, ignore it */
1053 if (q && q->handle == 0)
1056 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1057 if (tcm->tcm_handle) {
1058 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
1060 if (TC_H_MIN(tcm->tcm_handle))
1062 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
1063 goto create_n_graft;
1064 if (n->nlmsg_flags&NLM_F_EXCL)
1066 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1069 (p && check_loop(q, p, 0)))
1071 atomic_inc(&q->refcnt);
1075 goto create_n_graft;
1077 /* This magic test requires explanation.
1079 * We know, that some child q is already
1080 * attached to this parent and have choice:
1081 * either to change it or to create/graft new one.
1083 * 1. We are allowed to create/graft only
1084 * if CREATE and REPLACE flags are set.
1086 * 2. If EXCL is set, requestor wanted to say,
1087 * that qdisc tcm_handle is not expected
1088 * to exist, so that we choose create/graft too.
1090 * 3. The last case is when no flags are set.
1091 * Alas, it is sort of hole in API, we
1092 * cannot decide what to do unambiguously.
1093 * For now we select create/graft, if
1094 * user gave KIND, which does not match existing.
1096 if ((n->nlmsg_flags&NLM_F_CREATE) &&
1097 (n->nlmsg_flags&NLM_F_REPLACE) &&
1098 ((n->nlmsg_flags&NLM_F_EXCL) ||
1100 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1101 goto create_n_graft;
1105 if (!tcm->tcm_handle)
1107 q = qdisc_lookup(dev, tcm->tcm_handle);
1110 /* Change qdisc parameters */
1113 if (n->nlmsg_flags&NLM_F_EXCL)
1115 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1117 err = qdisc_change(q, tca);
1119 qdisc_notify(skb, n, clid, NULL, q);
1123 if (!(n->nlmsg_flags&NLM_F_CREATE))
1125 if (clid == TC_H_INGRESS)
1126 q = qdisc_create(dev, &dev->rx_queue,
1127 tcm->tcm_parent, tcm->tcm_parent,
1130 q = qdisc_create(dev, netdev_get_tx_queue(dev, 0),
1131 tcm->tcm_parent, tcm->tcm_handle,
1140 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1150 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1151 u32 pid, u32 seq, u16 flags, int event)
1154 struct nlmsghdr *nlh;
1155 unsigned char *b = skb_tail_pointer(skb);
1158 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1159 tcm = NLMSG_DATA(nlh);
1160 tcm->tcm_family = AF_UNSPEC;
1163 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1164 tcm->tcm_parent = clid;
1165 tcm->tcm_handle = q->handle;
1166 tcm->tcm_info = atomic_read(&q->refcnt);
1167 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1168 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1169 goto nla_put_failure;
1170 q->qstats.qlen = q->q.qlen;
1172 if (q->stab && qdisc_dump_stab(skb, q->stab) < 0)
1173 goto nla_put_failure;
1175 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1176 qdisc_root_sleeping_lock(q), &d) < 0)
1177 goto nla_put_failure;
1179 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1180 goto nla_put_failure;
1182 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1183 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
1184 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1185 goto nla_put_failure;
1187 if (gnet_stats_finish_copy(&d) < 0)
1188 goto nla_put_failure;
1190 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1199 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1200 u32 clid, struct Qdisc *old, struct Qdisc *new)
1202 struct sk_buff *skb;
1203 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1205 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1209 if (old && old->handle) {
1210 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
1214 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1219 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1226 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1228 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1231 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1232 struct netlink_callback *cb,
1233 int *q_idx_p, int s_q_idx)
1235 int ret = 0, q_idx = *q_idx_p;
1242 if (q_idx < s_q_idx) {
1245 if (!tc_qdisc_dump_ignore(q) &&
1246 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1247 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1251 list_for_each_entry(q, &root->list, list) {
1252 if (q_idx < s_q_idx) {
1256 if (!tc_qdisc_dump_ignore(q) &&
1257 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1258 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1271 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1273 struct net *net = sock_net(skb->sk);
1276 struct net_device *dev;
1278 if (net != &init_net)
1281 s_idx = cb->args[0];
1282 s_q_idx = q_idx = cb->args[1];
1283 read_lock(&dev_base_lock);
1285 for_each_netdev(&init_net, dev) {
1286 struct netdev_queue *dev_queue;
1294 dev_queue = netdev_get_tx_queue(dev, 0);
1295 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1298 dev_queue = &dev->rx_queue;
1299 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1307 read_unlock(&dev_base_lock);
1310 cb->args[1] = q_idx;
1317 /************************************************
1318 * Traffic classes manipulation. *
1319 ************************************************/
1323 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1325 struct net *net = sock_net(skb->sk);
1326 struct netdev_queue *dev_queue;
1327 struct tcmsg *tcm = NLMSG_DATA(n);
1328 struct nlattr *tca[TCA_MAX + 1];
1329 struct net_device *dev;
1330 struct Qdisc *q = NULL;
1331 const struct Qdisc_class_ops *cops;
1332 unsigned long cl = 0;
1333 unsigned long new_cl;
1334 u32 pid = tcm->tcm_parent;
1335 u32 clid = tcm->tcm_handle;
1336 u32 qid = TC_H_MAJ(clid);
1339 if (net != &init_net)
1342 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1345 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1350 parent == TC_H_UNSPEC - unspecified parent.
1351 parent == TC_H_ROOT - class is root, which has no parent.
1352 parent == X:0 - parent is root class.
1353 parent == X:Y - parent is a node in hierarchy.
1354 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1356 handle == 0:0 - generate handle from kernel pool.
1357 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1358 handle == X:Y - clear.
1359 handle == X:0 - root class.
1362 /* Step 1. Determine qdisc handle X:0 */
1364 dev_queue = netdev_get_tx_queue(dev, 0);
1365 if (pid != TC_H_ROOT) {
1366 u32 qid1 = TC_H_MAJ(pid);
1369 /* If both majors are known, they must be identical. */
1374 } else if (qid == 0)
1375 qid = dev_queue->qdisc_sleeping->handle;
1377 /* Now qid is genuine qdisc handle consistent
1378 both with parent and child.
1380 TC_H_MAJ(pid) still may be unspecified, complete it now.
1383 pid = TC_H_MAKE(qid, pid);
1386 qid = dev_queue->qdisc_sleeping->handle;
1389 /* OK. Locate qdisc */
1390 if ((q = qdisc_lookup(dev, qid)) == NULL)
1393 /* An check that it supports classes */
1394 cops = q->ops->cl_ops;
1398 /* Now try to get class */
1400 if (pid == TC_H_ROOT)
1403 clid = TC_H_MAKE(qid, clid);
1406 cl = cops->get(q, clid);
1410 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1413 switch (n->nlmsg_type) {
1416 if (n->nlmsg_flags&NLM_F_EXCL)
1420 err = cops->delete(q, cl);
1422 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1425 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1434 err = cops->change(q, clid, pid, tca, &new_cl);
1436 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1446 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1448 u32 pid, u32 seq, u16 flags, int event)
1451 struct nlmsghdr *nlh;
1452 unsigned char *b = skb_tail_pointer(skb);
1454 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1456 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1457 tcm = NLMSG_DATA(nlh);
1458 tcm->tcm_family = AF_UNSPEC;
1459 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1460 tcm->tcm_parent = q->handle;
1461 tcm->tcm_handle = q->handle;
1463 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1464 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1465 goto nla_put_failure;
1467 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1468 qdisc_root_sleeping_lock(q), &d) < 0)
1469 goto nla_put_failure;
1471 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1472 goto nla_put_failure;
1474 if (gnet_stats_finish_copy(&d) < 0)
1475 goto nla_put_failure;
1477 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1486 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1487 struct Qdisc *q, unsigned long cl, int event)
1489 struct sk_buff *skb;
1490 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1492 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1496 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1501 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1504 struct qdisc_dump_args
1506 struct qdisc_walker w;
1507 struct sk_buff *skb;
1508 struct netlink_callback *cb;
1511 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1513 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1515 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1516 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1519 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1520 struct tcmsg *tcm, struct netlink_callback *cb,
1523 struct qdisc_dump_args arg;
1525 if (tc_qdisc_dump_ignore(q) ||
1526 *t_p < s_t || !q->ops->cl_ops ||
1528 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1533 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1534 arg.w.fn = qdisc_class_dump;
1538 arg.w.skip = cb->args[1];
1540 q->ops->cl_ops->walk(q, &arg.w);
1541 cb->args[1] = arg.w.count;
1548 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1549 struct tcmsg *tcm, struct netlink_callback *cb,
1557 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1560 list_for_each_entry(q, &root->list, list) {
1561 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1568 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1570 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1571 struct net *net = sock_net(skb->sk);
1572 struct netdev_queue *dev_queue;
1573 struct net_device *dev;
1576 if (net != &init_net)
1579 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1581 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1587 dev_queue = netdev_get_tx_queue(dev, 0);
1588 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1591 dev_queue = &dev->rx_queue;
1592 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1602 /* Main classifier routine: scans classifier chain attached
1603 to this qdisc, (optionally) tests for protocol and asks
1604 specific classifiers.
1606 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1607 struct tcf_result *res)
1609 __be16 protocol = skb->protocol;
1612 for (; tp; tp = tp->next) {
1613 if ((tp->protocol == protocol ||
1614 tp->protocol == htons(ETH_P_ALL)) &&
1615 (err = tp->classify(skb, tp, res)) >= 0) {
1616 #ifdef CONFIG_NET_CLS_ACT
1617 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1618 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1625 EXPORT_SYMBOL(tc_classify_compat);
1627 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1628 struct tcf_result *res)
1632 #ifdef CONFIG_NET_CLS_ACT
1633 struct tcf_proto *otp = tp;
1636 protocol = skb->protocol;
1638 err = tc_classify_compat(skb, tp, res);
1639 #ifdef CONFIG_NET_CLS_ACT
1640 if (err == TC_ACT_RECLASSIFY) {
1641 u32 verd = G_TC_VERD(skb->tc_verd);
1644 if (verd++ >= MAX_REC_LOOP) {
1645 printk("rule prio %u protocol %02x reclassify loop, "
1647 tp->prio&0xffff, ntohs(tp->protocol));
1650 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1656 EXPORT_SYMBOL(tc_classify);
1658 void tcf_destroy(struct tcf_proto *tp)
1660 tp->ops->destroy(tp);
1661 module_put(tp->ops->owner);
1665 void tcf_destroy_chain(struct tcf_proto **fl)
1667 struct tcf_proto *tp;
1669 while ((tp = *fl) != NULL) {
1674 EXPORT_SYMBOL(tcf_destroy_chain);
1676 #ifdef CONFIG_PROC_FS
1677 static int psched_show(struct seq_file *seq, void *v)
1681 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1682 seq_printf(seq, "%08x %08x %08x %08x\n",
1683 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1685 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1690 static int psched_open(struct inode *inode, struct file *file)
1692 return single_open(file, psched_show, PDE(inode)->data);
1695 static const struct file_operations psched_fops = {
1696 .owner = THIS_MODULE,
1697 .open = psched_open,
1699 .llseek = seq_lseek,
1700 .release = single_release,
1704 static int __init pktsched_init(void)
1706 register_qdisc(&pfifo_qdisc_ops);
1707 register_qdisc(&bfifo_qdisc_ops);
1708 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1710 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1711 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1712 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1713 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1714 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1715 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1720 subsys_initcall(pktsched_init);