2 * net/sched/sch_api.c Packet scheduler API.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
18 #include <linux/module.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/string.h>
22 #include <linux/errno.h>
23 #include <linux/skbuff.h>
24 #include <linux/init.h>
25 #include <linux/proc_fs.h>
26 #include <linux/seq_file.h>
27 #include <linux/kmod.h>
28 #include <linux/list.h>
29 #include <linux/hrtimer.h>
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 sk_buff *oskb, struct nlmsghdr *n, u32 clid,
39 struct Qdisc *old, struct Qdisc *new);
40 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
41 struct Qdisc *q, unsigned long cl, int event);
48 This file consists of two interrelated parts:
50 1. queueing disciplines manager frontend.
51 2. traffic classes manager frontend.
53 Generally, queueing discipline ("qdisc") is a black box,
54 which is able to enqueue packets and to dequeue them (when
55 device is ready to send something) in order and at times
56 determined by algorithm hidden in it.
58 qdisc's are divided to two categories:
59 - "queues", which have no internal structure visible from outside.
60 - "schedulers", which split all the packets to "traffic classes",
61 using "packet classifiers" (look at cls_api.c)
63 In turn, classes may have child qdiscs (as rule, queues)
64 attached to them etc. etc. etc.
66 The goal of the routines in this file is to translate
67 information supplied by user in the form of handles
68 to more intelligible for kernel form, to make some sanity
69 checks and part of work, which is common to all qdiscs
70 and to provide rtnetlink notifications.
72 All real intelligent work is done inside qdisc modules.
76 Every discipline has two major routines: enqueue and dequeue.
80 dequeue usually returns a skb to send. It is allowed to return NULL,
81 but it does not mean that queue is empty, it just means that
82 discipline does not want to send anything this time.
83 Queue is really empty if q->q.qlen == 0.
84 For complicated disciplines with multiple queues q->q is not
85 real packet queue, but however q->q.qlen must be valid.
89 enqueue returns 0, if packet was enqueued successfully.
90 If packet (this one or another one) was dropped, it returns
92 NET_XMIT_DROP - this packet dropped
93 Expected action: do not backoff, but wait until queue will clear.
94 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
95 Expected action: backoff or ignore
96 NET_XMIT_POLICED - dropped by police.
97 Expected action: backoff or error to real-time apps.
103 like dequeue but without removing a packet from the queue
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->peek == NULL) {
151 if (qops->dequeue == NULL) {
152 qops->peek = noop_qdisc_ops.peek;
158 if (qops->dequeue == NULL)
159 qops->dequeue = noop_qdisc_ops.dequeue;
165 write_unlock(&qdisc_mod_lock);
168 EXPORT_SYMBOL(register_qdisc);
170 int unregister_qdisc(struct Qdisc_ops *qops)
172 struct Qdisc_ops *q, **qp;
175 write_lock(&qdisc_mod_lock);
176 for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next)
184 write_unlock(&qdisc_mod_lock);
187 EXPORT_SYMBOL(unregister_qdisc);
189 /* We know handle. Find qdisc among all qdisc's attached to device
190 (root qdisc, all its children, children of children etc.)
193 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
197 if (!(root->flags & TCQ_F_BUILTIN) &&
198 root->handle == handle)
201 list_for_each_entry(q, &root->list, list) {
202 if (q->handle == handle)
208 static void qdisc_list_add(struct Qdisc *q)
210 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
211 list_add_tail(&q->list, &qdisc_dev(q)->qdisc->list);
214 void qdisc_list_del(struct Qdisc *q)
216 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
219 EXPORT_SYMBOL(qdisc_list_del);
221 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
225 q = qdisc_match_from_root(dev->qdisc, handle);
229 q = qdisc_match_from_root(dev->rx_queue.qdisc_sleeping, handle);
234 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
238 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
242 cl = cops->get(p, classid);
246 leaf = cops->leaf(p, cl);
251 /* Find queueing discipline by name */
253 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
255 struct Qdisc_ops *q = NULL;
258 read_lock(&qdisc_mod_lock);
259 for (q = qdisc_base; q; q = q->next) {
260 if (nla_strcmp(kind, q->id) == 0) {
261 if (!try_module_get(q->owner))
266 read_unlock(&qdisc_mod_lock);
271 static struct qdisc_rate_table *qdisc_rtab_list;
273 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
275 struct qdisc_rate_table *rtab;
277 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
278 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
284 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
285 nla_len(tab) != TC_RTAB_SIZE)
288 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
292 memcpy(rtab->data, nla_data(tab), 1024);
293 rtab->next = qdisc_rtab_list;
294 qdisc_rtab_list = rtab;
298 EXPORT_SYMBOL(qdisc_get_rtab);
300 void qdisc_put_rtab(struct qdisc_rate_table *tab)
302 struct qdisc_rate_table *rtab, **rtabp;
304 if (!tab || --tab->refcnt)
307 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
315 EXPORT_SYMBOL(qdisc_put_rtab);
317 static LIST_HEAD(qdisc_stab_list);
318 static DEFINE_SPINLOCK(qdisc_stab_lock);
320 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
321 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
322 [TCA_STAB_DATA] = { .type = NLA_BINARY },
325 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
327 struct nlattr *tb[TCA_STAB_MAX + 1];
328 struct qdisc_size_table *stab;
329 struct tc_sizespec *s;
330 unsigned int tsize = 0;
334 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
337 if (!tb[TCA_STAB_BASE])
338 return ERR_PTR(-EINVAL);
340 s = nla_data(tb[TCA_STAB_BASE]);
343 if (!tb[TCA_STAB_DATA])
344 return ERR_PTR(-EINVAL);
345 tab = nla_data(tb[TCA_STAB_DATA]);
346 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
349 if (!s || tsize != s->tsize || (!tab && tsize > 0))
350 return ERR_PTR(-EINVAL);
352 spin_lock(&qdisc_stab_lock);
354 list_for_each_entry(stab, &qdisc_stab_list, list) {
355 if (memcmp(&stab->szopts, s, sizeof(*s)))
357 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
360 spin_unlock(&qdisc_stab_lock);
364 spin_unlock(&qdisc_stab_lock);
366 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
368 return ERR_PTR(-ENOMEM);
373 memcpy(stab->data, tab, tsize * sizeof(u16));
375 spin_lock(&qdisc_stab_lock);
376 list_add_tail(&stab->list, &qdisc_stab_list);
377 spin_unlock(&qdisc_stab_lock);
382 void qdisc_put_stab(struct qdisc_size_table *tab)
387 spin_lock(&qdisc_stab_lock);
389 if (--tab->refcnt == 0) {
390 list_del(&tab->list);
394 spin_unlock(&qdisc_stab_lock);
396 EXPORT_SYMBOL(qdisc_put_stab);
398 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
402 nest = nla_nest_start(skb, TCA_STAB);
404 goto nla_put_failure;
405 NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts);
406 nla_nest_end(skb, nest);
414 void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab)
418 pkt_len = skb->len + stab->szopts.overhead;
419 if (unlikely(!stab->szopts.tsize))
422 slot = pkt_len + stab->szopts.cell_align;
423 if (unlikely(slot < 0))
426 slot >>= stab->szopts.cell_log;
427 if (likely(slot < stab->szopts.tsize))
428 pkt_len = stab->data[slot];
430 pkt_len = stab->data[stab->szopts.tsize - 1] *
431 (slot / stab->szopts.tsize) +
432 stab->data[slot % stab->szopts.tsize];
434 pkt_len <<= stab->szopts.size_log;
436 if (unlikely(pkt_len < 1))
438 qdisc_skb_cb(skb)->pkt_len = pkt_len;
440 EXPORT_SYMBOL(qdisc_calculate_pkt_len);
442 void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
444 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
446 "%s: %s qdisc %X: is non-work-conserving?\n",
447 txt, qdisc->ops->id, qdisc->handle >> 16);
448 qdisc->flags |= TCQ_F_WARN_NONWC;
451 EXPORT_SYMBOL(qdisc_warn_nonwc);
453 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
455 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
458 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
459 __netif_schedule(qdisc_root(wd->qdisc));
461 return HRTIMER_NORESTART;
464 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
466 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
467 wd->timer.function = qdisc_watchdog;
470 EXPORT_SYMBOL(qdisc_watchdog_init);
472 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
476 if (test_bit(__QDISC_STATE_DEACTIVATED,
477 &qdisc_root_sleeping(wd->qdisc)->state))
480 wd->qdisc->flags |= TCQ_F_THROTTLED;
481 time = ktime_set(0, 0);
482 time = ktime_add_ns(time, PSCHED_TICKS2NS(expires));
483 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
485 EXPORT_SYMBOL(qdisc_watchdog_schedule);
487 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
489 hrtimer_cancel(&wd->timer);
490 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
492 EXPORT_SYMBOL(qdisc_watchdog_cancel);
494 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
496 unsigned int size = n * sizeof(struct hlist_head), i;
497 struct hlist_head *h;
499 if (size <= PAGE_SIZE)
500 h = kmalloc(size, GFP_KERNEL);
502 h = (struct hlist_head *)
503 __get_free_pages(GFP_KERNEL, get_order(size));
506 for (i = 0; i < n; i++)
507 INIT_HLIST_HEAD(&h[i]);
512 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
514 unsigned int size = n * sizeof(struct hlist_head);
516 if (size <= PAGE_SIZE)
519 free_pages((unsigned long)h, get_order(size));
522 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
524 struct Qdisc_class_common *cl;
525 struct hlist_node *n, *next;
526 struct hlist_head *nhash, *ohash;
527 unsigned int nsize, nmask, osize;
530 /* Rehash when load factor exceeds 0.75 */
531 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
533 nsize = clhash->hashsize * 2;
535 nhash = qdisc_class_hash_alloc(nsize);
539 ohash = clhash->hash;
540 osize = clhash->hashsize;
543 for (i = 0; i < osize; i++) {
544 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
545 h = qdisc_class_hash(cl->classid, nmask);
546 hlist_add_head(&cl->hnode, &nhash[h]);
549 clhash->hash = nhash;
550 clhash->hashsize = nsize;
551 clhash->hashmask = nmask;
552 sch_tree_unlock(sch);
554 qdisc_class_hash_free(ohash, osize);
556 EXPORT_SYMBOL(qdisc_class_hash_grow);
558 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
560 unsigned int size = 4;
562 clhash->hash = qdisc_class_hash_alloc(size);
563 if (clhash->hash == NULL)
565 clhash->hashsize = size;
566 clhash->hashmask = size - 1;
567 clhash->hashelems = 0;
570 EXPORT_SYMBOL(qdisc_class_hash_init);
572 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
574 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
576 EXPORT_SYMBOL(qdisc_class_hash_destroy);
578 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
579 struct Qdisc_class_common *cl)
583 INIT_HLIST_NODE(&cl->hnode);
584 h = qdisc_class_hash(cl->classid, clhash->hashmask);
585 hlist_add_head(&cl->hnode, &clhash->hash[h]);
588 EXPORT_SYMBOL(qdisc_class_hash_insert);
590 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
591 struct Qdisc_class_common *cl)
593 hlist_del(&cl->hnode);
596 EXPORT_SYMBOL(qdisc_class_hash_remove);
598 /* Allocate an unique handle from space managed by kernel */
600 static u32 qdisc_alloc_handle(struct net_device *dev)
603 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
606 autohandle += TC_H_MAKE(0x10000U, 0);
607 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
608 autohandle = TC_H_MAKE(0x80000000U, 0);
609 } while (qdisc_lookup(dev, autohandle) && --i > 0);
611 return i>0 ? autohandle : 0;
614 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
616 const struct Qdisc_class_ops *cops;
622 while ((parentid = sch->parent)) {
623 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
626 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
628 WARN_ON(parentid != TC_H_ROOT);
631 cops = sch->ops->cl_ops;
632 if (cops->qlen_notify) {
633 cl = cops->get(sch, parentid);
634 cops->qlen_notify(sch, cl);
640 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
642 static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid,
643 struct Qdisc *old, struct Qdisc *new)
646 qdisc_notify(skb, n, clid, old, new);
652 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
655 * When appropriate send a netlink notification using 'skb'
658 * On success, destroy old qdisc.
661 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
662 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
663 struct Qdisc *new, struct Qdisc *old)
665 struct Qdisc *q = old;
668 if (parent == NULL) {
669 unsigned int i, num_q, ingress;
672 num_q = dev->num_tx_queues;
673 if ((q && q->flags & TCQ_F_INGRESS) ||
674 (new && new->flags & TCQ_F_INGRESS)) {
679 if (dev->flags & IFF_UP)
682 if (new && new->ops->attach) {
683 new->ops->attach(new);
687 for (i = 0; i < num_q; i++) {
688 struct netdev_queue *dev_queue = &dev->rx_queue;
691 dev_queue = netdev_get_tx_queue(dev, i);
693 old = dev_graft_qdisc(dev_queue, new);
695 atomic_inc(&new->refcnt);
702 notify_and_destroy(skb, n, classid, dev->qdisc, new);
703 if (new && !new->ops->attach)
704 atomic_inc(&new->refcnt);
705 dev->qdisc = new ? : &noop_qdisc;
707 notify_and_destroy(skb, n, classid, old, new);
710 if (dev->flags & IFF_UP)
713 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
716 if (cops && cops->graft) {
717 unsigned long cl = cops->get(parent, classid);
719 err = cops->graft(parent, cl, new, &old);
720 cops->put(parent, cl);
725 notify_and_destroy(skb, n, classid, old, new);
730 /* lockdep annotation is needed for ingress; egress gets it only for name */
731 static struct lock_class_key qdisc_tx_lock;
732 static struct lock_class_key qdisc_rx_lock;
735 Allocate and initialize new qdisc.
737 Parameters are passed via opt.
740 static struct Qdisc *
741 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
742 struct Qdisc *p, u32 parent, u32 handle,
743 struct nlattr **tca, int *errp)
746 struct nlattr *kind = tca[TCA_KIND];
748 struct Qdisc_ops *ops;
749 struct qdisc_size_table *stab;
751 ops = qdisc_lookup_ops(kind);
752 #ifdef CONFIG_MODULES
753 if (ops == NULL && kind != NULL) {
755 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
756 /* We dropped the RTNL semaphore in order to
757 * perform the module load. So, even if we
758 * succeeded in loading the module we have to
759 * tell the caller to replay the request. We
760 * indicate this using -EAGAIN.
761 * We replay the request because the device may
762 * go away in the mean time.
765 request_module("sch_%s", name);
767 ops = qdisc_lookup_ops(kind);
769 /* We will try again qdisc_lookup_ops,
770 * so don't keep a reference.
772 module_put(ops->owner);
784 sch = qdisc_alloc(dev_queue, ops);
790 sch->parent = parent;
792 if (handle == TC_H_INGRESS) {
793 sch->flags |= TCQ_F_INGRESS;
794 handle = TC_H_MAKE(TC_H_INGRESS, 0);
795 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
798 handle = qdisc_alloc_handle(dev);
803 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
806 sch->handle = handle;
808 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
810 stab = qdisc_get_stab(tca[TCA_STAB]);
818 spinlock_t *root_lock;
821 if (sch->flags & TCQ_F_MQROOT)
824 if ((sch->parent != TC_H_ROOT) &&
825 !(sch->flags & TCQ_F_INGRESS) &&
826 (!p || !(p->flags & TCQ_F_MQROOT)))
827 root_lock = qdisc_root_sleeping_lock(sch);
829 root_lock = qdisc_lock(sch);
831 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
832 root_lock, tca[TCA_RATE]);
843 kfree((char *) sch - sch->padded);
845 module_put(ops->owner);
852 * Any broken qdiscs that would require a ops->reset() here?
853 * The qdisc was never in action so it shouldn't be necessary.
855 qdisc_put_stab(sch->stab);
861 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
863 struct qdisc_size_table *stab = NULL;
866 if (tca[TCA_OPTIONS]) {
867 if (sch->ops->change == NULL)
869 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
875 stab = qdisc_get_stab(tca[TCA_STAB]);
877 return PTR_ERR(stab);
880 qdisc_put_stab(sch->stab);
884 /* NB: ignores errors from replace_estimator
885 because change can't be undone. */
886 if (sch->flags & TCQ_F_MQROOT)
888 gen_replace_estimator(&sch->bstats, &sch->rate_est,
889 qdisc_root_sleeping_lock(sch),
896 struct check_loop_arg
898 struct qdisc_walker w;
903 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
905 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
907 struct check_loop_arg arg;
909 if (q->ops->cl_ops == NULL)
912 arg.w.stop = arg.w.skip = arg.w.count = 0;
913 arg.w.fn = check_loop_fn;
916 q->ops->cl_ops->walk(q, &arg.w);
917 return arg.w.stop ? -ELOOP : 0;
921 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
924 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
925 struct check_loop_arg *arg = (struct check_loop_arg *)w;
927 leaf = cops->leaf(q, cl);
929 if (leaf == arg->p || arg->depth > 7)
931 return check_loop(leaf, arg->p, arg->depth + 1);
940 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
942 struct net *net = sock_net(skb->sk);
943 struct tcmsg *tcm = NLMSG_DATA(n);
944 struct nlattr *tca[TCA_MAX + 1];
945 struct net_device *dev;
946 u32 clid = tcm->tcm_parent;
947 struct Qdisc *q = NULL;
948 struct Qdisc *p = NULL;
951 if (!net_eq(net, &init_net))
954 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
957 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
962 if (clid != TC_H_ROOT) {
963 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
964 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
966 q = qdisc_leaf(p, clid);
967 } else { /* ingress */
968 q = dev->rx_queue.qdisc_sleeping;
976 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
979 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
983 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
986 if (n->nlmsg_type == RTM_DELQDISC) {
991 if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0)
994 qdisc_notify(skb, n, clid, NULL, q);
1000 Create/change qdisc.
1003 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1005 struct net *net = sock_net(skb->sk);
1007 struct nlattr *tca[TCA_MAX + 1];
1008 struct net_device *dev;
1010 struct Qdisc *q, *p;
1013 if (!net_eq(net, &init_net))
1017 /* Reinit, just in case something touches this. */
1018 tcm = NLMSG_DATA(n);
1019 clid = tcm->tcm_parent;
1022 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1025 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1030 if (clid != TC_H_ROOT) {
1031 if (clid != TC_H_INGRESS) {
1032 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
1034 q = qdisc_leaf(p, clid);
1035 } else { /*ingress */
1036 q = dev->rx_queue.qdisc_sleeping;
1042 /* It may be default qdisc, ignore it */
1043 if (q && q->handle == 0)
1046 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1047 if (tcm->tcm_handle) {
1048 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
1050 if (TC_H_MIN(tcm->tcm_handle))
1052 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
1053 goto create_n_graft;
1054 if (n->nlmsg_flags&NLM_F_EXCL)
1056 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1059 (p && check_loop(q, p, 0)))
1061 atomic_inc(&q->refcnt);
1065 goto create_n_graft;
1067 /* This magic test requires explanation.
1069 * We know, that some child q is already
1070 * attached to this parent and have choice:
1071 * either to change it or to create/graft new one.
1073 * 1. We are allowed to create/graft only
1074 * if CREATE and REPLACE flags are set.
1076 * 2. If EXCL is set, requestor wanted to say,
1077 * that qdisc tcm_handle is not expected
1078 * to exist, so that we choose create/graft too.
1080 * 3. The last case is when no flags are set.
1081 * Alas, it is sort of hole in API, we
1082 * cannot decide what to do unambiguously.
1083 * For now we select create/graft, if
1084 * user gave KIND, which does not match existing.
1086 if ((n->nlmsg_flags&NLM_F_CREATE) &&
1087 (n->nlmsg_flags&NLM_F_REPLACE) &&
1088 ((n->nlmsg_flags&NLM_F_EXCL) ||
1090 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1091 goto create_n_graft;
1095 if (!tcm->tcm_handle)
1097 q = qdisc_lookup(dev, tcm->tcm_handle);
1100 /* Change qdisc parameters */
1103 if (n->nlmsg_flags&NLM_F_EXCL)
1105 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1107 err = qdisc_change(q, tca);
1109 qdisc_notify(skb, n, clid, NULL, q);
1113 if (!(n->nlmsg_flags&NLM_F_CREATE))
1115 if (clid == TC_H_INGRESS)
1116 q = qdisc_create(dev, &dev->rx_queue, p,
1117 tcm->tcm_parent, tcm->tcm_parent,
1120 struct netdev_queue *dev_queue;
1122 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1123 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1125 dev_queue = p->dev_queue;
1127 dev_queue = netdev_get_tx_queue(dev, 0);
1129 q = qdisc_create(dev, dev_queue, p,
1130 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->bstats, &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_eq(net, &init_net))
1281 s_idx = cb->args[0];
1282 s_q_idx = q_idx = cb->args[1];
1286 for_each_netdev_rcu(&init_net, dev) {
1287 struct netdev_queue *dev_queue;
1295 if (tc_dump_qdisc_root(dev->qdisc, 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)
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 tcmsg *tcm = NLMSG_DATA(n);
1327 struct nlattr *tca[TCA_MAX + 1];
1328 struct net_device *dev;
1329 struct Qdisc *q = NULL;
1330 const struct Qdisc_class_ops *cops;
1331 unsigned long cl = 0;
1332 unsigned long new_cl;
1333 u32 pid = tcm->tcm_parent;
1334 u32 clid = tcm->tcm_handle;
1335 u32 qid = TC_H_MAJ(clid);
1338 if (!net_eq(net, &init_net))
1341 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1344 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1349 parent == TC_H_UNSPEC - unspecified parent.
1350 parent == TC_H_ROOT - class is root, which has no parent.
1351 parent == X:0 - parent is root class.
1352 parent == X:Y - parent is a node in hierarchy.
1353 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1355 handle == 0:0 - generate handle from kernel pool.
1356 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1357 handle == X:Y - clear.
1358 handle == X:0 - root class.
1361 /* Step 1. Determine qdisc handle X:0 */
1363 if (pid != TC_H_ROOT) {
1364 u32 qid1 = TC_H_MAJ(pid);
1367 /* If both majors are known, they must be identical. */
1372 } else if (qid == 0)
1373 qid = dev->qdisc->handle;
1375 /* Now qid is genuine qdisc handle consistent
1376 both with parent and child.
1378 TC_H_MAJ(pid) still may be unspecified, complete it now.
1381 pid = TC_H_MAKE(qid, pid);
1384 qid = dev->qdisc->handle;
1387 /* OK. Locate qdisc */
1388 if ((q = qdisc_lookup(dev, qid)) == NULL)
1391 /* An check that it supports classes */
1392 cops = q->ops->cl_ops;
1396 /* Now try to get class */
1398 if (pid == TC_H_ROOT)
1401 clid = TC_H_MAKE(qid, clid);
1404 cl = cops->get(q, clid);
1408 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1411 switch (n->nlmsg_type) {
1414 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);
1436 err = cops->change(q, clid, pid, tca, &new_cl);
1438 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1448 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1450 u32 pid, u32 seq, u16 flags, int event)
1453 struct nlmsghdr *nlh;
1454 unsigned char *b = skb_tail_pointer(skb);
1456 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1458 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1459 tcm = NLMSG_DATA(nlh);
1460 tcm->tcm_family = AF_UNSPEC;
1463 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1464 tcm->tcm_parent = q->handle;
1465 tcm->tcm_handle = q->handle;
1467 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1468 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1469 goto nla_put_failure;
1471 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1472 qdisc_root_sleeping_lock(q), &d) < 0)
1473 goto nla_put_failure;
1475 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1476 goto nla_put_failure;
1478 if (gnet_stats_finish_copy(&d) < 0)
1479 goto nla_put_failure;
1481 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1490 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1491 struct Qdisc *q, unsigned long cl, int event)
1493 struct sk_buff *skb;
1494 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1496 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1500 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1505 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1508 struct qdisc_dump_args
1510 struct qdisc_walker w;
1511 struct sk_buff *skb;
1512 struct netlink_callback *cb;
1515 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1517 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1519 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1520 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1523 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1524 struct tcmsg *tcm, struct netlink_callback *cb,
1527 struct qdisc_dump_args arg;
1529 if (tc_qdisc_dump_ignore(q) ||
1530 *t_p < s_t || !q->ops->cl_ops ||
1532 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1537 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1538 arg.w.fn = qdisc_class_dump;
1542 arg.w.skip = cb->args[1];
1544 q->ops->cl_ops->walk(q, &arg.w);
1545 cb->args[1] = arg.w.count;
1552 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1553 struct tcmsg *tcm, struct netlink_callback *cb,
1561 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1564 list_for_each_entry(q, &root->list, list) {
1565 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1572 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1574 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1575 struct net *net = sock_net(skb->sk);
1576 struct netdev_queue *dev_queue;
1577 struct net_device *dev;
1580 if (!net_eq(net, &init_net))
1583 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1585 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1591 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1594 dev_queue = &dev->rx_queue;
1595 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
1605 /* Main classifier routine: scans classifier chain attached
1606 to this qdisc, (optionally) tests for protocol and asks
1607 specific classifiers.
1609 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1610 struct tcf_result *res)
1612 __be16 protocol = skb->protocol;
1615 for (; tp; tp = tp->next) {
1616 if ((tp->protocol == protocol ||
1617 tp->protocol == htons(ETH_P_ALL)) &&
1618 (err = tp->classify(skb, tp, res)) >= 0) {
1619 #ifdef CONFIG_NET_CLS_ACT
1620 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1621 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1628 EXPORT_SYMBOL(tc_classify_compat);
1630 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1631 struct tcf_result *res)
1635 #ifdef CONFIG_NET_CLS_ACT
1636 struct tcf_proto *otp = tp;
1639 protocol = skb->protocol;
1641 err = tc_classify_compat(skb, tp, res);
1642 #ifdef CONFIG_NET_CLS_ACT
1643 if (err == TC_ACT_RECLASSIFY) {
1644 u32 verd = G_TC_VERD(skb->tc_verd);
1647 if (verd++ >= MAX_REC_LOOP) {
1648 printk("rule prio %u protocol %02x reclassify loop, "
1650 tp->prio&0xffff, ntohs(tp->protocol));
1653 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1659 EXPORT_SYMBOL(tc_classify);
1661 void tcf_destroy(struct tcf_proto *tp)
1663 tp->ops->destroy(tp);
1664 module_put(tp->ops->owner);
1668 void tcf_destroy_chain(struct tcf_proto **fl)
1670 struct tcf_proto *tp;
1672 while ((tp = *fl) != NULL) {
1677 EXPORT_SYMBOL(tcf_destroy_chain);
1679 #ifdef CONFIG_PROC_FS
1680 static int psched_show(struct seq_file *seq, void *v)
1684 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1685 seq_printf(seq, "%08x %08x %08x %08x\n",
1686 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1688 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1693 static int psched_open(struct inode *inode, struct file *file)
1695 return single_open(file, psched_show, PDE(inode)->data);
1698 static const struct file_operations psched_fops = {
1699 .owner = THIS_MODULE,
1700 .open = psched_open,
1702 .llseek = seq_lseek,
1703 .release = single_release,
1707 static int __init pktsched_init(void)
1709 register_qdisc(&pfifo_qdisc_ops);
1710 register_qdisc(&bfifo_qdisc_ops);
1711 register_qdisc(&pfifo_head_drop_qdisc_ops);
1712 register_qdisc(&mq_qdisc_ops);
1713 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1715 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1716 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1717 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1718 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1719 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1720 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1725 subsys_initcall(pktsched_init);