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>
32 #include <linux/hashtable.h>
34 #include <net/net_namespace.h>
36 #include <net/netlink.h>
37 #include <net/pkt_sched.h>
39 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
40 struct nlmsghdr *n, u32 clid,
41 struct Qdisc *old, struct Qdisc *new);
42 static int tclass_notify(struct net *net, struct sk_buff *oskb,
43 struct nlmsghdr *n, struct Qdisc *q,
44 unsigned long cl, int event);
51 This file consists of two interrelated parts:
53 1. queueing disciplines manager frontend.
54 2. traffic classes manager frontend.
56 Generally, queueing discipline ("qdisc") is a black box,
57 which is able to enqueue packets and to dequeue them (when
58 device is ready to send something) in order and at times
59 determined by algorithm hidden in it.
61 qdisc's are divided to two categories:
62 - "queues", which have no internal structure visible from outside.
63 - "schedulers", which split all the packets to "traffic classes",
64 using "packet classifiers" (look at cls_api.c)
66 In turn, classes may have child qdiscs (as rule, queues)
67 attached to them etc. etc. etc.
69 The goal of the routines in this file is to translate
70 information supplied by user in the form of handles
71 to more intelligible for kernel form, to make some sanity
72 checks and part of work, which is common to all qdiscs
73 and to provide rtnetlink notifications.
75 All real intelligent work is done inside qdisc modules.
79 Every discipline has two major routines: enqueue and dequeue.
83 dequeue usually returns a skb to send. It is allowed to return NULL,
84 but it does not mean that queue is empty, it just means that
85 discipline does not want to send anything this time.
86 Queue is really empty if q->q.qlen == 0.
87 For complicated disciplines with multiple queues q->q is not
88 real packet queue, but however q->q.qlen must be valid.
92 enqueue returns 0, if packet was enqueued successfully.
93 If packet (this one or another one) was dropped, it returns
95 NET_XMIT_DROP - this packet dropped
96 Expected action: do not backoff, but wait until queue will clear.
97 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
98 Expected action: backoff or ignore
104 like dequeue but without removing a packet from the queue
108 returns qdisc to initial state: purge all buffers, clear all
109 timers, counters (except for statistics) etc.
113 initializes newly created qdisc.
117 destroys resources allocated by init and during lifetime of qdisc.
121 changes qdisc parameters.
124 /* Protects list of registered TC modules. It is pure SMP lock. */
125 static DEFINE_RWLOCK(qdisc_mod_lock);
128 /************************************************
129 * Queueing disciplines manipulation. *
130 ************************************************/
133 /* The list of all installed queueing disciplines. */
135 static struct Qdisc_ops *qdisc_base;
137 /* Register/unregister queueing discipline */
139 int register_qdisc(struct Qdisc_ops *qops)
141 struct Qdisc_ops *q, **qp;
144 write_lock(&qdisc_mod_lock);
145 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
146 if (!strcmp(qops->id, q->id))
149 if (qops->enqueue == NULL)
150 qops->enqueue = noop_qdisc_ops.enqueue;
151 if (qops->peek == NULL) {
152 if (qops->dequeue == NULL)
153 qops->peek = noop_qdisc_ops.peek;
157 if (qops->dequeue == NULL)
158 qops->dequeue = noop_qdisc_ops.dequeue;
161 const struct Qdisc_class_ops *cops = qops->cl_ops;
163 if (!(cops->get && cops->put && cops->walk && cops->leaf))
166 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf))
174 write_unlock(&qdisc_mod_lock);
181 EXPORT_SYMBOL(register_qdisc);
183 int unregister_qdisc(struct Qdisc_ops *qops)
185 struct Qdisc_ops *q, **qp;
188 write_lock(&qdisc_mod_lock);
189 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
197 write_unlock(&qdisc_mod_lock);
200 EXPORT_SYMBOL(unregister_qdisc);
202 /* Get default qdisc if not otherwise specified */
203 void qdisc_get_default(char *name, size_t len)
205 read_lock(&qdisc_mod_lock);
206 strlcpy(name, default_qdisc_ops->id, len);
207 read_unlock(&qdisc_mod_lock);
210 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
212 struct Qdisc_ops *q = NULL;
214 for (q = qdisc_base; q; q = q->next) {
215 if (!strcmp(name, q->id)) {
216 if (!try_module_get(q->owner))
225 /* Set new default qdisc to use */
226 int qdisc_set_default(const char *name)
228 const struct Qdisc_ops *ops;
230 if (!capable(CAP_NET_ADMIN))
233 write_lock(&qdisc_mod_lock);
234 ops = qdisc_lookup_default(name);
236 /* Not found, drop lock and try to load module */
237 write_unlock(&qdisc_mod_lock);
238 request_module("sch_%s", name);
239 write_lock(&qdisc_mod_lock);
241 ops = qdisc_lookup_default(name);
245 /* Set new default */
246 module_put(default_qdisc_ops->owner);
247 default_qdisc_ops = ops;
249 write_unlock(&qdisc_mod_lock);
251 return ops ? 0 : -ENOENT;
254 /* We know handle. Find qdisc among all qdisc's attached to device
255 * (root qdisc, all its children, children of children etc.)
256 * Note: caller either uses rtnl or rcu_read_lock()
259 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
263 if (!qdisc_dev(root))
264 return (root->handle == handle ? root : NULL);
266 if (!(root->flags & TCQ_F_BUILTIN) &&
267 root->handle == handle)
270 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) {
271 if (q->handle == handle)
277 void qdisc_hash_add(struct Qdisc *q)
279 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
280 struct Qdisc *root = qdisc_dev(q)->qdisc;
282 WARN_ON_ONCE(root == &noop_qdisc);
284 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
287 EXPORT_SYMBOL(qdisc_hash_add);
289 void qdisc_hash_del(struct Qdisc *q)
291 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
293 hash_del_rcu(&q->hash);
296 EXPORT_SYMBOL(qdisc_hash_del);
298 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
302 q = qdisc_match_from_root(dev->qdisc, handle);
306 if (dev_ingress_queue(dev))
307 q = qdisc_match_from_root(
308 dev_ingress_queue(dev)->qdisc_sleeping,
314 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
318 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
322 cl = cops->get(p, classid);
326 leaf = cops->leaf(p, cl);
331 /* Find queueing discipline by name */
333 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
335 struct Qdisc_ops *q = NULL;
338 read_lock(&qdisc_mod_lock);
339 for (q = qdisc_base; q; q = q->next) {
340 if (nla_strcmp(kind, q->id) == 0) {
341 if (!try_module_get(q->owner))
346 read_unlock(&qdisc_mod_lock);
351 /* The linklayer setting were not transferred from iproute2, in older
352 * versions, and the rate tables lookup systems have been dropped in
353 * the kernel. To keep backward compatible with older iproute2 tc
354 * utils, we detect the linklayer setting by detecting if the rate
355 * table were modified.
357 * For linklayer ATM table entries, the rate table will be aligned to
358 * 48 bytes, thus some table entries will contain the same value. The
359 * mpu (min packet unit) is also encoded into the old rate table, thus
360 * starting from the mpu, we find low and high table entries for
361 * mapping this cell. If these entries contain the same value, when
362 * the rate tables have been modified for linklayer ATM.
364 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
365 * and then roundup to the next cell, calc the table entry one below,
368 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
370 int low = roundup(r->mpu, 48);
371 int high = roundup(low+1, 48);
372 int cell_low = low >> r->cell_log;
373 int cell_high = (high >> r->cell_log) - 1;
375 /* rtab is too inaccurate at rates > 100Mbit/s */
376 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
377 pr_debug("TC linklayer: Giving up ATM detection\n");
378 return TC_LINKLAYER_ETHERNET;
381 if ((cell_high > cell_low) && (cell_high < 256)
382 && (rtab[cell_low] == rtab[cell_high])) {
383 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
384 cell_low, cell_high, rtab[cell_high]);
385 return TC_LINKLAYER_ATM;
387 return TC_LINKLAYER_ETHERNET;
390 static struct qdisc_rate_table *qdisc_rtab_list;
392 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
395 struct qdisc_rate_table *rtab;
397 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
398 nla_len(tab) != TC_RTAB_SIZE)
401 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
402 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
403 !memcmp(&rtab->data, nla_data(tab), 1024)) {
409 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
413 memcpy(rtab->data, nla_data(tab), 1024);
414 if (r->linklayer == TC_LINKLAYER_UNAWARE)
415 r->linklayer = __detect_linklayer(r, rtab->data);
416 rtab->next = qdisc_rtab_list;
417 qdisc_rtab_list = rtab;
421 EXPORT_SYMBOL(qdisc_get_rtab);
423 void qdisc_put_rtab(struct qdisc_rate_table *tab)
425 struct qdisc_rate_table *rtab, **rtabp;
427 if (!tab || --tab->refcnt)
430 for (rtabp = &qdisc_rtab_list;
431 (rtab = *rtabp) != NULL;
432 rtabp = &rtab->next) {
440 EXPORT_SYMBOL(qdisc_put_rtab);
442 static LIST_HEAD(qdisc_stab_list);
443 static DEFINE_SPINLOCK(qdisc_stab_lock);
445 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
446 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
447 [TCA_STAB_DATA] = { .type = NLA_BINARY },
450 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
452 struct nlattr *tb[TCA_STAB_MAX + 1];
453 struct qdisc_size_table *stab;
454 struct tc_sizespec *s;
455 unsigned int tsize = 0;
459 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
462 if (!tb[TCA_STAB_BASE])
463 return ERR_PTR(-EINVAL);
465 s = nla_data(tb[TCA_STAB_BASE]);
468 if (!tb[TCA_STAB_DATA])
469 return ERR_PTR(-EINVAL);
470 tab = nla_data(tb[TCA_STAB_DATA]);
471 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
474 if (tsize != s->tsize || (!tab && tsize > 0))
475 return ERR_PTR(-EINVAL);
477 spin_lock(&qdisc_stab_lock);
479 list_for_each_entry(stab, &qdisc_stab_list, list) {
480 if (memcmp(&stab->szopts, s, sizeof(*s)))
482 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
485 spin_unlock(&qdisc_stab_lock);
489 spin_unlock(&qdisc_stab_lock);
491 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
493 return ERR_PTR(-ENOMEM);
498 memcpy(stab->data, tab, tsize * sizeof(u16));
500 spin_lock(&qdisc_stab_lock);
501 list_add_tail(&stab->list, &qdisc_stab_list);
502 spin_unlock(&qdisc_stab_lock);
507 static void stab_kfree_rcu(struct rcu_head *head)
509 kfree(container_of(head, struct qdisc_size_table, rcu));
512 void qdisc_put_stab(struct qdisc_size_table *tab)
517 spin_lock(&qdisc_stab_lock);
519 if (--tab->refcnt == 0) {
520 list_del(&tab->list);
521 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
524 spin_unlock(&qdisc_stab_lock);
526 EXPORT_SYMBOL(qdisc_put_stab);
528 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
532 nest = nla_nest_start(skb, TCA_STAB);
534 goto nla_put_failure;
535 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
536 goto nla_put_failure;
537 nla_nest_end(skb, nest);
545 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
546 const struct qdisc_size_table *stab)
550 pkt_len = skb->len + stab->szopts.overhead;
551 if (unlikely(!stab->szopts.tsize))
554 slot = pkt_len + stab->szopts.cell_align;
555 if (unlikely(slot < 0))
558 slot >>= stab->szopts.cell_log;
559 if (likely(slot < stab->szopts.tsize))
560 pkt_len = stab->data[slot];
562 pkt_len = stab->data[stab->szopts.tsize - 1] *
563 (slot / stab->szopts.tsize) +
564 stab->data[slot % stab->szopts.tsize];
566 pkt_len <<= stab->szopts.size_log;
568 if (unlikely(pkt_len < 1))
570 qdisc_skb_cb(skb)->pkt_len = pkt_len;
572 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
574 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
576 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
577 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
578 txt, qdisc->ops->id, qdisc->handle >> 16);
579 qdisc->flags |= TCQ_F_WARN_NONWC;
582 EXPORT_SYMBOL(qdisc_warn_nonwc);
584 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
586 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
590 __netif_schedule(qdisc_root(wd->qdisc));
593 return HRTIMER_NORESTART;
596 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
598 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
599 wd->timer.function = qdisc_watchdog;
602 EXPORT_SYMBOL(qdisc_watchdog_init);
604 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
606 if (test_bit(__QDISC_STATE_DEACTIVATED,
607 &qdisc_root_sleeping(wd->qdisc)->state))
610 if (wd->last_expires == expires)
613 wd->last_expires = expires;
614 hrtimer_start(&wd->timer,
615 ns_to_ktime(expires),
616 HRTIMER_MODE_ABS_PINNED);
618 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
620 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
622 hrtimer_cancel(&wd->timer);
624 EXPORT_SYMBOL(qdisc_watchdog_cancel);
626 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
628 unsigned int size = n * sizeof(struct hlist_head), i;
629 struct hlist_head *h;
631 if (size <= PAGE_SIZE)
632 h = kmalloc(size, GFP_KERNEL);
634 h = (struct hlist_head *)
635 __get_free_pages(GFP_KERNEL, get_order(size));
638 for (i = 0; i < n; i++)
639 INIT_HLIST_HEAD(&h[i]);
644 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
646 unsigned int size = n * sizeof(struct hlist_head);
648 if (size <= PAGE_SIZE)
651 free_pages((unsigned long)h, get_order(size));
654 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
656 struct Qdisc_class_common *cl;
657 struct hlist_node *next;
658 struct hlist_head *nhash, *ohash;
659 unsigned int nsize, nmask, osize;
662 /* Rehash when load factor exceeds 0.75 */
663 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
665 nsize = clhash->hashsize * 2;
667 nhash = qdisc_class_hash_alloc(nsize);
671 ohash = clhash->hash;
672 osize = clhash->hashsize;
675 for (i = 0; i < osize; i++) {
676 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
677 h = qdisc_class_hash(cl->classid, nmask);
678 hlist_add_head(&cl->hnode, &nhash[h]);
681 clhash->hash = nhash;
682 clhash->hashsize = nsize;
683 clhash->hashmask = nmask;
684 sch_tree_unlock(sch);
686 qdisc_class_hash_free(ohash, osize);
688 EXPORT_SYMBOL(qdisc_class_hash_grow);
690 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
692 unsigned int size = 4;
694 clhash->hash = qdisc_class_hash_alloc(size);
695 if (clhash->hash == NULL)
697 clhash->hashsize = size;
698 clhash->hashmask = size - 1;
699 clhash->hashelems = 0;
702 EXPORT_SYMBOL(qdisc_class_hash_init);
704 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
706 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
708 EXPORT_SYMBOL(qdisc_class_hash_destroy);
710 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
711 struct Qdisc_class_common *cl)
715 INIT_HLIST_NODE(&cl->hnode);
716 h = qdisc_class_hash(cl->classid, clhash->hashmask);
717 hlist_add_head(&cl->hnode, &clhash->hash[h]);
720 EXPORT_SYMBOL(qdisc_class_hash_insert);
722 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
723 struct Qdisc_class_common *cl)
725 hlist_del(&cl->hnode);
728 EXPORT_SYMBOL(qdisc_class_hash_remove);
730 /* Allocate an unique handle from space managed by kernel
731 * Possible range is [8000-FFFF]:0000 (0x8000 values)
733 static u32 qdisc_alloc_handle(struct net_device *dev)
736 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
739 autohandle += TC_H_MAKE(0x10000U, 0);
740 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
741 autohandle = TC_H_MAKE(0x80000000U, 0);
742 if (!qdisc_lookup(dev, autohandle))
750 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n,
753 const struct Qdisc_class_ops *cops;
758 if (n == 0 && len == 0)
760 drops = max_t(int, n, 0);
762 while ((parentid = sch->parent)) {
763 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
766 if (sch->flags & TCQ_F_NOPARENT)
768 /* TODO: perform the search on a per txq basis */
769 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
771 WARN_ON_ONCE(parentid != TC_H_ROOT);
774 cops = sch->ops->cl_ops;
775 if (cops->qlen_notify) {
776 cl = cops->get(sch, parentid);
777 cops->qlen_notify(sch, cl);
781 sch->qstats.backlog -= len;
782 __qdisc_qstats_drop(sch, drops);
786 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
788 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
789 struct nlmsghdr *n, u32 clid,
790 struct Qdisc *old, struct Qdisc *new)
793 qdisc_notify(net, skb, n, clid, old, new);
799 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
802 * When appropriate send a netlink notification using 'skb'
805 * On success, destroy old qdisc.
808 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
809 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
810 struct Qdisc *new, struct Qdisc *old)
812 struct Qdisc *q = old;
813 struct net *net = dev_net(dev);
816 if (parent == NULL) {
817 unsigned int i, num_q, ingress;
820 num_q = dev->num_tx_queues;
821 if ((q && q->flags & TCQ_F_INGRESS) ||
822 (new && new->flags & TCQ_F_INGRESS)) {
825 if (!dev_ingress_queue(dev))
829 if (dev->flags & IFF_UP)
832 if (new && new->ops->attach)
835 for (i = 0; i < num_q; i++) {
836 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
839 dev_queue = netdev_get_tx_queue(dev, i);
841 old = dev_graft_qdisc(dev_queue, new);
843 atomic_inc(&new->refcnt);
851 notify_and_destroy(net, skb, n, classid,
853 if (new && !new->ops->attach)
854 atomic_inc(&new->refcnt);
855 dev->qdisc = new ? : &noop_qdisc;
857 if (new && new->ops->attach)
858 new->ops->attach(new);
860 notify_and_destroy(net, skb, n, classid, old, new);
863 if (dev->flags & IFF_UP)
866 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
869 if (cops && cops->graft) {
870 unsigned long cl = cops->get(parent, classid);
872 err = cops->graft(parent, cl, new, &old);
873 cops->put(parent, cl);
878 notify_and_destroy(net, skb, n, classid, old, new);
883 /* lockdep annotation is needed for ingress; egress gets it only for name */
884 static struct lock_class_key qdisc_tx_lock;
885 static struct lock_class_key qdisc_rx_lock;
888 Allocate and initialize new qdisc.
890 Parameters are passed via opt.
893 static struct Qdisc *qdisc_create(struct net_device *dev,
894 struct netdev_queue *dev_queue,
895 struct Qdisc *p, u32 parent, u32 handle,
896 struct nlattr **tca, int *errp)
899 struct nlattr *kind = tca[TCA_KIND];
901 struct Qdisc_ops *ops;
902 struct qdisc_size_table *stab;
904 ops = qdisc_lookup_ops(kind);
905 #ifdef CONFIG_MODULES
906 if (ops == NULL && kind != NULL) {
908 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
909 /* We dropped the RTNL semaphore in order to
910 * perform the module load. So, even if we
911 * succeeded in loading the module we have to
912 * tell the caller to replay the request. We
913 * indicate this using -EAGAIN.
914 * We replay the request because the device may
915 * go away in the mean time.
918 request_module("sch_%s", name);
920 ops = qdisc_lookup_ops(kind);
922 /* We will try again qdisc_lookup_ops,
923 * so don't keep a reference.
925 module_put(ops->owner);
937 sch = qdisc_alloc(dev_queue, ops);
943 sch->parent = parent;
945 if (handle == TC_H_INGRESS) {
946 sch->flags |= TCQ_F_INGRESS;
947 handle = TC_H_MAKE(TC_H_INGRESS, 0);
948 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
951 handle = qdisc_alloc_handle(dev);
956 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
957 if (!netif_is_multiqueue(dev))
958 sch->flags |= TCQ_F_ONETXQUEUE;
961 sch->handle = handle;
963 /* This exist to keep backward compatible with a userspace
964 * loophole, what allowed userspace to get IFF_NO_QUEUE
965 * facility on older kernels by setting tx_queue_len=0 (prior
966 * to qdisc init), and then forgot to reinit tx_queue_len
967 * before again attaching a qdisc.
969 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
970 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
971 netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
974 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
975 if (qdisc_is_percpu_stats(sch)) {
977 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
978 if (!sch->cpu_bstats)
981 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
982 if (!sch->cpu_qstats)
987 stab = qdisc_get_stab(tca[TCA_STAB]);
992 rcu_assign_pointer(sch->stab, stab);
998 if (sch->flags & TCQ_F_MQROOT)
1001 if ((sch->parent != TC_H_ROOT) &&
1002 !(sch->flags & TCQ_F_INGRESS) &&
1003 (!p || !(p->flags & TCQ_F_MQROOT)))
1004 running = qdisc_root_sleeping_running(sch);
1006 running = &sch->running;
1008 err = gen_new_estimator(&sch->bstats,
1018 qdisc_hash_add(sch);
1024 kfree((char *) sch - sch->padded);
1026 module_put(ops->owner);
1032 free_percpu(sch->cpu_bstats);
1033 free_percpu(sch->cpu_qstats);
1035 * Any broken qdiscs that would require a ops->reset() here?
1036 * The qdisc was never in action so it shouldn't be necessary.
1038 qdisc_put_stab(rtnl_dereference(sch->stab));
1044 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
1046 struct qdisc_size_table *ostab, *stab = NULL;
1049 if (tca[TCA_OPTIONS]) {
1050 if (sch->ops->change == NULL)
1052 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1057 if (tca[TCA_STAB]) {
1058 stab = qdisc_get_stab(tca[TCA_STAB]);
1060 return PTR_ERR(stab);
1063 ostab = rtnl_dereference(sch->stab);
1064 rcu_assign_pointer(sch->stab, stab);
1065 qdisc_put_stab(ostab);
1067 if (tca[TCA_RATE]) {
1068 /* NB: ignores errors from replace_estimator
1069 because change can't be undone. */
1070 if (sch->flags & TCQ_F_MQROOT)
1072 gen_replace_estimator(&sch->bstats,
1076 qdisc_root_sleeping_running(sch),
1083 struct check_loop_arg {
1084 struct qdisc_walker w;
1089 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1090 struct qdisc_walker *w);
1092 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1094 struct check_loop_arg arg;
1096 if (q->ops->cl_ops == NULL)
1099 arg.w.stop = arg.w.skip = arg.w.count = 0;
1100 arg.w.fn = check_loop_fn;
1103 q->ops->cl_ops->walk(q, &arg.w);
1104 return arg.w.stop ? -ELOOP : 0;
1108 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1111 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1112 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1114 leaf = cops->leaf(q, cl);
1116 if (leaf == arg->p || arg->depth > 7)
1118 return check_loop(leaf, arg->p, arg->depth + 1);
1127 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1129 struct net *net = sock_net(skb->sk);
1130 struct tcmsg *tcm = nlmsg_data(n);
1131 struct nlattr *tca[TCA_MAX + 1];
1132 struct net_device *dev;
1134 struct Qdisc *q = NULL;
1135 struct Qdisc *p = NULL;
1138 if ((n->nlmsg_type != RTM_GETQDISC) &&
1139 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1142 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1146 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1150 clid = tcm->tcm_parent;
1152 if (clid != TC_H_ROOT) {
1153 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1154 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1157 q = qdisc_leaf(p, clid);
1158 } else if (dev_ingress_queue(dev)) {
1159 q = dev_ingress_queue(dev)->qdisc_sleeping;
1167 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1170 q = qdisc_lookup(dev, tcm->tcm_handle);
1175 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1178 if (n->nlmsg_type == RTM_DELQDISC) {
1183 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1187 qdisc_notify(net, skb, n, clid, NULL, q);
1193 * Create/change qdisc.
1196 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1198 struct net *net = sock_net(skb->sk);
1200 struct nlattr *tca[TCA_MAX + 1];
1201 struct net_device *dev;
1203 struct Qdisc *q, *p;
1206 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1210 /* Reinit, just in case something touches this. */
1211 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1215 tcm = nlmsg_data(n);
1216 clid = tcm->tcm_parent;
1219 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1225 if (clid != TC_H_ROOT) {
1226 if (clid != TC_H_INGRESS) {
1227 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1230 q = qdisc_leaf(p, clid);
1231 } else if (dev_ingress_queue_create(dev)) {
1232 q = dev_ingress_queue(dev)->qdisc_sleeping;
1238 /* It may be default qdisc, ignore it */
1239 if (q && q->handle == 0)
1242 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1243 if (tcm->tcm_handle) {
1244 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1246 if (TC_H_MIN(tcm->tcm_handle))
1248 q = qdisc_lookup(dev, tcm->tcm_handle);
1250 goto create_n_graft;
1251 if (n->nlmsg_flags & NLM_F_EXCL)
1253 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1256 (p && check_loop(q, p, 0)))
1258 atomic_inc(&q->refcnt);
1262 goto create_n_graft;
1264 /* This magic test requires explanation.
1266 * We know, that some child q is already
1267 * attached to this parent and have choice:
1268 * either to change it or to create/graft new one.
1270 * 1. We are allowed to create/graft only
1271 * if CREATE and REPLACE flags are set.
1273 * 2. If EXCL is set, requestor wanted to say,
1274 * that qdisc tcm_handle is not expected
1275 * to exist, so that we choose create/graft too.
1277 * 3. The last case is when no flags are set.
1278 * Alas, it is sort of hole in API, we
1279 * cannot decide what to do unambiguously.
1280 * For now we select create/graft, if
1281 * user gave KIND, which does not match existing.
1283 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1284 (n->nlmsg_flags & NLM_F_REPLACE) &&
1285 ((n->nlmsg_flags & NLM_F_EXCL) ||
1287 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1288 goto create_n_graft;
1292 if (!tcm->tcm_handle)
1294 q = qdisc_lookup(dev, tcm->tcm_handle);
1297 /* Change qdisc parameters */
1300 if (n->nlmsg_flags & NLM_F_EXCL)
1302 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1304 err = qdisc_change(q, tca);
1306 qdisc_notify(net, skb, n, clid, NULL, q);
1310 if (!(n->nlmsg_flags & NLM_F_CREATE))
1312 if (clid == TC_H_INGRESS) {
1313 if (dev_ingress_queue(dev))
1314 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1315 tcm->tcm_parent, tcm->tcm_parent,
1320 struct netdev_queue *dev_queue;
1322 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1323 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1325 dev_queue = p->dev_queue;
1327 dev_queue = netdev_get_tx_queue(dev, 0);
1329 q = qdisc_create(dev, dev_queue, p,
1330 tcm->tcm_parent, tcm->tcm_handle,
1340 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1350 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1351 u32 portid, u32 seq, u16 flags, int event)
1353 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
1354 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
1356 struct nlmsghdr *nlh;
1357 unsigned char *b = skb_tail_pointer(skb);
1359 struct qdisc_size_table *stab;
1363 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1365 goto out_nlmsg_trim;
1366 tcm = nlmsg_data(nlh);
1367 tcm->tcm_family = AF_UNSPEC;
1370 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1371 tcm->tcm_parent = clid;
1372 tcm->tcm_handle = q->handle;
1373 tcm->tcm_info = atomic_read(&q->refcnt);
1374 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1375 goto nla_put_failure;
1376 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1377 goto nla_put_failure;
1380 stab = rtnl_dereference(q->stab);
1381 if (stab && qdisc_dump_stab(skb, stab) < 0)
1382 goto nla_put_failure;
1384 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1385 NULL, &d, TCA_PAD) < 0)
1386 goto nla_put_failure;
1388 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1389 goto nla_put_failure;
1391 if (qdisc_is_percpu_stats(q)) {
1392 cpu_bstats = q->cpu_bstats;
1393 cpu_qstats = q->cpu_qstats;
1396 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
1397 &d, cpu_bstats, &q->bstats) < 0 ||
1398 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
1399 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
1400 goto nla_put_failure;
1402 if (gnet_stats_finish_copy(&d) < 0)
1403 goto nla_put_failure;
1405 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1414 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1416 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1419 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1420 struct nlmsghdr *n, u32 clid,
1421 struct Qdisc *old, struct Qdisc *new)
1423 struct sk_buff *skb;
1424 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1426 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1430 if (old && !tc_qdisc_dump_ignore(old)) {
1431 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1432 0, RTM_DELQDISC) < 0)
1435 if (new && !tc_qdisc_dump_ignore(new)) {
1436 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1437 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1442 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1443 n->nlmsg_flags & NLM_F_ECHO);
1450 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1451 struct netlink_callback *cb,
1452 int *q_idx_p, int s_q_idx, bool recur)
1454 int ret = 0, q_idx = *q_idx_p;
1462 if (q_idx < s_q_idx) {
1465 if (!tc_qdisc_dump_ignore(q) &&
1466 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1467 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1473 /* If dumping singletons, there is no qdisc_dev(root) and the singleton
1474 * itself has already been dumped.
1476 * If we've already dumped the top-level (ingress) qdisc above and the global
1477 * qdisc hashtable, we don't want to hit it again
1479 if (!qdisc_dev(root) || !recur)
1482 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1483 if (q_idx < s_q_idx) {
1487 if (!tc_qdisc_dump_ignore(q) &&
1488 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1489 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1503 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1505 struct net *net = sock_net(skb->sk);
1508 struct net_device *dev;
1510 s_idx = cb->args[0];
1511 s_q_idx = q_idx = cb->args[1];
1515 for_each_netdev(net, dev) {
1516 struct netdev_queue *dev_queue;
1524 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1528 dev_queue = dev_ingress_queue(dev);
1530 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1531 &q_idx, s_q_idx, false) < 0)
1540 cb->args[1] = q_idx;
1547 /************************************************
1548 * Traffic classes manipulation. *
1549 ************************************************/
1553 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1555 struct net *net = sock_net(skb->sk);
1556 struct tcmsg *tcm = nlmsg_data(n);
1557 struct nlattr *tca[TCA_MAX + 1];
1558 struct net_device *dev;
1559 struct Qdisc *q = NULL;
1560 const struct Qdisc_class_ops *cops;
1561 unsigned long cl = 0;
1562 unsigned long new_cl;
1568 if ((n->nlmsg_type != RTM_GETTCLASS) &&
1569 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1572 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1576 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1581 parent == TC_H_UNSPEC - unspecified parent.
1582 parent == TC_H_ROOT - class is root, which has no parent.
1583 parent == X:0 - parent is root class.
1584 parent == X:Y - parent is a node in hierarchy.
1585 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1587 handle == 0:0 - generate handle from kernel pool.
1588 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1589 handle == X:Y - clear.
1590 handle == X:0 - root class.
1593 /* Step 1. Determine qdisc handle X:0 */
1595 portid = tcm->tcm_parent;
1596 clid = tcm->tcm_handle;
1597 qid = TC_H_MAJ(clid);
1599 if (portid != TC_H_ROOT) {
1600 u32 qid1 = TC_H_MAJ(portid);
1603 /* If both majors are known, they must be identical. */
1608 } else if (qid == 0)
1609 qid = dev->qdisc->handle;
1611 /* Now qid is genuine qdisc handle consistent
1612 * both with parent and child.
1614 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1617 portid = TC_H_MAKE(qid, portid);
1620 qid = dev->qdisc->handle;
1623 /* OK. Locate qdisc */
1624 q = qdisc_lookup(dev, qid);
1628 /* An check that it supports classes */
1629 cops = q->ops->cl_ops;
1633 /* Now try to get class */
1635 if (portid == TC_H_ROOT)
1638 clid = TC_H_MAKE(qid, clid);
1641 cl = cops->get(q, clid);
1645 if (n->nlmsg_type != RTM_NEWTCLASS ||
1646 !(n->nlmsg_flags & NLM_F_CREATE))
1649 switch (n->nlmsg_type) {
1652 if (n->nlmsg_flags & NLM_F_EXCL)
1658 err = cops->delete(q, cl);
1660 tclass_notify(net, skb, n, q, cl,
1664 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1675 err = cops->change(q, clid, portid, tca, &new_cl);
1677 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1687 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1689 u32 portid, u32 seq, u16 flags, int event)
1692 struct nlmsghdr *nlh;
1693 unsigned char *b = skb_tail_pointer(skb);
1695 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1698 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1700 goto out_nlmsg_trim;
1701 tcm = nlmsg_data(nlh);
1702 tcm->tcm_family = AF_UNSPEC;
1705 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1706 tcm->tcm_parent = q->handle;
1707 tcm->tcm_handle = q->handle;
1709 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1710 goto nla_put_failure;
1711 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1712 goto nla_put_failure;
1714 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1715 NULL, &d, TCA_PAD) < 0)
1716 goto nla_put_failure;
1718 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1719 goto nla_put_failure;
1721 if (gnet_stats_finish_copy(&d) < 0)
1722 goto nla_put_failure;
1724 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1733 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1734 struct nlmsghdr *n, struct Qdisc *q,
1735 unsigned long cl, int event)
1737 struct sk_buff *skb;
1738 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1740 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1744 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1749 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1750 n->nlmsg_flags & NLM_F_ECHO);
1753 struct qdisc_dump_args {
1754 struct qdisc_walker w;
1755 struct sk_buff *skb;
1756 struct netlink_callback *cb;
1759 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
1760 struct qdisc_walker *arg)
1762 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1764 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1765 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1769 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1770 struct tcmsg *tcm, struct netlink_callback *cb,
1773 struct qdisc_dump_args arg;
1775 if (tc_qdisc_dump_ignore(q) ||
1776 *t_p < s_t || !q->ops->cl_ops ||
1778 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1783 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1784 arg.w.fn = qdisc_class_dump;
1788 arg.w.skip = cb->args[1];
1790 q->ops->cl_ops->walk(q, &arg.w);
1791 cb->args[1] = arg.w.count;
1798 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1799 struct tcmsg *tcm, struct netlink_callback *cb,
1808 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1811 if (!qdisc_dev(root))
1814 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1815 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1822 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1824 struct tcmsg *tcm = nlmsg_data(cb->nlh);
1825 struct net *net = sock_net(skb->sk);
1826 struct netdev_queue *dev_queue;
1827 struct net_device *dev;
1830 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1832 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1839 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1842 dev_queue = dev_ingress_queue(dev);
1844 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1855 /* Main classifier routine: scans classifier chain attached
1856 * to this qdisc, (optionally) tests for protocol and asks
1857 * specific classifiers.
1859 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1860 struct tcf_result *res, bool compat_mode)
1862 __be16 protocol = tc_skb_protocol(skb);
1863 #ifdef CONFIG_NET_CLS_ACT
1864 const int max_reclassify_loop = 4;
1865 const struct tcf_proto *old_tp = tp;
1870 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1873 if (tp->protocol != protocol &&
1874 tp->protocol != htons(ETH_P_ALL))
1877 err = tp->classify(skb, tp, res);
1878 #ifdef CONFIG_NET_CLS_ACT
1879 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode))
1886 return TC_ACT_UNSPEC; /* signal: continue lookup */
1887 #ifdef CONFIG_NET_CLS_ACT
1889 if (unlikely(limit++ >= max_reclassify_loop)) {
1890 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
1891 tp->q->ops->id, tp->prio & 0xffff,
1892 ntohs(tp->protocol));
1897 protocol = tc_skb_protocol(skb);
1901 EXPORT_SYMBOL(tc_classify);
1903 bool tcf_destroy(struct tcf_proto *tp, bool force)
1905 if (tp->ops->destroy(tp, force)) {
1906 module_put(tp->ops->owner);
1914 void tcf_destroy_chain(struct tcf_proto __rcu **fl)
1916 struct tcf_proto *tp;
1918 while ((tp = rtnl_dereference(*fl)) != NULL) {
1919 RCU_INIT_POINTER(*fl, tp->next);
1920 tcf_destroy(tp, true);
1923 EXPORT_SYMBOL(tcf_destroy_chain);
1925 #ifdef CONFIG_PROC_FS
1926 static int psched_show(struct seq_file *seq, void *v)
1928 seq_printf(seq, "%08x %08x %08x %08x\n",
1929 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1931 (u32)NSEC_PER_SEC / hrtimer_resolution);
1936 static int psched_open(struct inode *inode, struct file *file)
1938 return single_open(file, psched_show, NULL);
1941 static const struct file_operations psched_fops = {
1942 .owner = THIS_MODULE,
1943 .open = psched_open,
1945 .llseek = seq_lseek,
1946 .release = single_release,
1949 static int __net_init psched_net_init(struct net *net)
1951 struct proc_dir_entry *e;
1953 e = proc_create("psched", 0, net->proc_net, &psched_fops);
1960 static void __net_exit psched_net_exit(struct net *net)
1962 remove_proc_entry("psched", net->proc_net);
1965 static int __net_init psched_net_init(struct net *net)
1970 static void __net_exit psched_net_exit(struct net *net)
1975 static struct pernet_operations psched_net_ops = {
1976 .init = psched_net_init,
1977 .exit = psched_net_exit,
1980 static int __init pktsched_init(void)
1984 err = register_pernet_subsys(&psched_net_ops);
1986 pr_err("pktsched_init: "
1987 "cannot initialize per netns operations\n");
1991 register_qdisc(&pfifo_fast_ops);
1992 register_qdisc(&pfifo_qdisc_ops);
1993 register_qdisc(&bfifo_qdisc_ops);
1994 register_qdisc(&pfifo_head_drop_qdisc_ops);
1995 register_qdisc(&mq_qdisc_ops);
1996 register_qdisc(&noqueue_qdisc_ops);
1998 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1999 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
2000 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2002 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
2003 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
2004 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2010 subsys_initcall(pktsched_init);