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 #ifdef CONFIG_NET_SCH_DEFAULT
255 /* Set default value from kernel config */
256 static int __init sch_default_qdisc(void)
258 return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
260 late_initcall(sch_default_qdisc);
263 /* We know handle. Find qdisc among all qdisc's attached to device
264 * (root qdisc, all its children, children of children etc.)
265 * Note: caller either uses rtnl or rcu_read_lock()
268 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
272 if (!qdisc_dev(root))
273 return (root->handle == handle ? root : NULL);
275 if (!(root->flags & TCQ_F_BUILTIN) &&
276 root->handle == handle)
279 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) {
280 if (q->handle == handle)
286 void qdisc_hash_add(struct Qdisc *q, bool invisible)
288 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
289 struct Qdisc *root = qdisc_dev(q)->qdisc;
291 WARN_ON_ONCE(root == &noop_qdisc);
293 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
295 q->flags |= TCQ_F_INVISIBLE;
298 EXPORT_SYMBOL(qdisc_hash_add);
300 void qdisc_hash_del(struct Qdisc *q)
302 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
304 hash_del_rcu(&q->hash);
307 EXPORT_SYMBOL(qdisc_hash_del);
309 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
313 q = qdisc_match_from_root(dev->qdisc, handle);
317 if (dev_ingress_queue(dev))
318 q = qdisc_match_from_root(
319 dev_ingress_queue(dev)->qdisc_sleeping,
325 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
329 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
333 cl = cops->get(p, classid);
337 leaf = cops->leaf(p, cl);
342 /* Find queueing discipline by name */
344 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
346 struct Qdisc_ops *q = NULL;
349 read_lock(&qdisc_mod_lock);
350 for (q = qdisc_base; q; q = q->next) {
351 if (nla_strcmp(kind, q->id) == 0) {
352 if (!try_module_get(q->owner))
357 read_unlock(&qdisc_mod_lock);
362 /* The linklayer setting were not transferred from iproute2, in older
363 * versions, and the rate tables lookup systems have been dropped in
364 * the kernel. To keep backward compatible with older iproute2 tc
365 * utils, we detect the linklayer setting by detecting if the rate
366 * table were modified.
368 * For linklayer ATM table entries, the rate table will be aligned to
369 * 48 bytes, thus some table entries will contain the same value. The
370 * mpu (min packet unit) is also encoded into the old rate table, thus
371 * starting from the mpu, we find low and high table entries for
372 * mapping this cell. If these entries contain the same value, when
373 * the rate tables have been modified for linklayer ATM.
375 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
376 * and then roundup to the next cell, calc the table entry one below,
379 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
381 int low = roundup(r->mpu, 48);
382 int high = roundup(low+1, 48);
383 int cell_low = low >> r->cell_log;
384 int cell_high = (high >> r->cell_log) - 1;
386 /* rtab is too inaccurate at rates > 100Mbit/s */
387 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
388 pr_debug("TC linklayer: Giving up ATM detection\n");
389 return TC_LINKLAYER_ETHERNET;
392 if ((cell_high > cell_low) && (cell_high < 256)
393 && (rtab[cell_low] == rtab[cell_high])) {
394 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
395 cell_low, cell_high, rtab[cell_high]);
396 return TC_LINKLAYER_ATM;
398 return TC_LINKLAYER_ETHERNET;
401 static struct qdisc_rate_table *qdisc_rtab_list;
403 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
406 struct qdisc_rate_table *rtab;
408 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
409 nla_len(tab) != TC_RTAB_SIZE)
412 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
413 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
414 !memcmp(&rtab->data, nla_data(tab), 1024)) {
420 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
424 memcpy(rtab->data, nla_data(tab), 1024);
425 if (r->linklayer == TC_LINKLAYER_UNAWARE)
426 r->linklayer = __detect_linklayer(r, rtab->data);
427 rtab->next = qdisc_rtab_list;
428 qdisc_rtab_list = rtab;
432 EXPORT_SYMBOL(qdisc_get_rtab);
434 void qdisc_put_rtab(struct qdisc_rate_table *tab)
436 struct qdisc_rate_table *rtab, **rtabp;
438 if (!tab || --tab->refcnt)
441 for (rtabp = &qdisc_rtab_list;
442 (rtab = *rtabp) != NULL;
443 rtabp = &rtab->next) {
451 EXPORT_SYMBOL(qdisc_put_rtab);
453 static LIST_HEAD(qdisc_stab_list);
455 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
456 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
457 [TCA_STAB_DATA] = { .type = NLA_BINARY },
460 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
462 struct nlattr *tb[TCA_STAB_MAX + 1];
463 struct qdisc_size_table *stab;
464 struct tc_sizespec *s;
465 unsigned int tsize = 0;
469 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy, NULL);
472 if (!tb[TCA_STAB_BASE])
473 return ERR_PTR(-EINVAL);
475 s = nla_data(tb[TCA_STAB_BASE]);
478 if (!tb[TCA_STAB_DATA])
479 return ERR_PTR(-EINVAL);
480 tab = nla_data(tb[TCA_STAB_DATA]);
481 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
484 if (tsize != s->tsize || (!tab && tsize > 0))
485 return ERR_PTR(-EINVAL);
487 list_for_each_entry(stab, &qdisc_stab_list, list) {
488 if (memcmp(&stab->szopts, s, sizeof(*s)))
490 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
496 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
498 return ERR_PTR(-ENOMEM);
503 memcpy(stab->data, tab, tsize * sizeof(u16));
505 list_add_tail(&stab->list, &qdisc_stab_list);
510 static void stab_kfree_rcu(struct rcu_head *head)
512 kfree(container_of(head, struct qdisc_size_table, rcu));
515 void qdisc_put_stab(struct qdisc_size_table *tab)
520 if (--tab->refcnt == 0) {
521 list_del(&tab->list);
522 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
525 EXPORT_SYMBOL(qdisc_put_stab);
527 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
531 nest = nla_nest_start(skb, TCA_STAB);
533 goto nla_put_failure;
534 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
535 goto nla_put_failure;
536 nla_nest_end(skb, nest);
544 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
545 const struct qdisc_size_table *stab)
549 pkt_len = skb->len + stab->szopts.overhead;
550 if (unlikely(!stab->szopts.tsize))
553 slot = pkt_len + stab->szopts.cell_align;
554 if (unlikely(slot < 0))
557 slot >>= stab->szopts.cell_log;
558 if (likely(slot < stab->szopts.tsize))
559 pkt_len = stab->data[slot];
561 pkt_len = stab->data[stab->szopts.tsize - 1] *
562 (slot / stab->szopts.tsize) +
563 stab->data[slot % stab->szopts.tsize];
565 pkt_len <<= stab->szopts.size_log;
567 if (unlikely(pkt_len < 1))
569 qdisc_skb_cb(skb)->pkt_len = pkt_len;
571 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
573 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
575 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
576 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
577 txt, qdisc->ops->id, qdisc->handle >> 16);
578 qdisc->flags |= TCQ_F_WARN_NONWC;
581 EXPORT_SYMBOL(qdisc_warn_nonwc);
583 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
585 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
589 __netif_schedule(qdisc_root(wd->qdisc));
592 return HRTIMER_NORESTART;
595 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
597 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
598 wd->timer.function = qdisc_watchdog;
601 EXPORT_SYMBOL(qdisc_watchdog_init);
603 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
605 if (test_bit(__QDISC_STATE_DEACTIVATED,
606 &qdisc_root_sleeping(wd->qdisc)->state))
609 if (wd->last_expires == expires)
612 wd->last_expires = expires;
613 hrtimer_start(&wd->timer,
614 ns_to_ktime(expires),
615 HRTIMER_MODE_ABS_PINNED);
617 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
619 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
621 hrtimer_cancel(&wd->timer);
623 EXPORT_SYMBOL(qdisc_watchdog_cancel);
625 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
627 unsigned int size = n * sizeof(struct hlist_head), i;
628 struct hlist_head *h;
630 if (size <= PAGE_SIZE)
631 h = kmalloc(size, GFP_KERNEL);
633 h = (struct hlist_head *)
634 __get_free_pages(GFP_KERNEL, get_order(size));
637 for (i = 0; i < n; i++)
638 INIT_HLIST_HEAD(&h[i]);
643 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
645 unsigned int size = n * sizeof(struct hlist_head);
647 if (size <= PAGE_SIZE)
650 free_pages((unsigned long)h, get_order(size));
653 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
655 struct Qdisc_class_common *cl;
656 struct hlist_node *next;
657 struct hlist_head *nhash, *ohash;
658 unsigned int nsize, nmask, osize;
661 /* Rehash when load factor exceeds 0.75 */
662 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
664 nsize = clhash->hashsize * 2;
666 nhash = qdisc_class_hash_alloc(nsize);
670 ohash = clhash->hash;
671 osize = clhash->hashsize;
674 for (i = 0; i < osize; i++) {
675 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
676 h = qdisc_class_hash(cl->classid, nmask);
677 hlist_add_head(&cl->hnode, &nhash[h]);
680 clhash->hash = nhash;
681 clhash->hashsize = nsize;
682 clhash->hashmask = nmask;
683 sch_tree_unlock(sch);
685 qdisc_class_hash_free(ohash, osize);
687 EXPORT_SYMBOL(qdisc_class_hash_grow);
689 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
691 unsigned int size = 4;
693 clhash->hash = qdisc_class_hash_alloc(size);
694 if (clhash->hash == NULL)
696 clhash->hashsize = size;
697 clhash->hashmask = size - 1;
698 clhash->hashelems = 0;
701 EXPORT_SYMBOL(qdisc_class_hash_init);
703 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
705 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
707 EXPORT_SYMBOL(qdisc_class_hash_destroy);
709 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
710 struct Qdisc_class_common *cl)
714 INIT_HLIST_NODE(&cl->hnode);
715 h = qdisc_class_hash(cl->classid, clhash->hashmask);
716 hlist_add_head(&cl->hnode, &clhash->hash[h]);
719 EXPORT_SYMBOL(qdisc_class_hash_insert);
721 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
722 struct Qdisc_class_common *cl)
724 hlist_del(&cl->hnode);
727 EXPORT_SYMBOL(qdisc_class_hash_remove);
729 /* Allocate an unique handle from space managed by kernel
730 * Possible range is [8000-FFFF]:0000 (0x8000 values)
732 static u32 qdisc_alloc_handle(struct net_device *dev)
735 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
738 autohandle += TC_H_MAKE(0x10000U, 0);
739 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
740 autohandle = TC_H_MAKE(0x80000000U, 0);
741 if (!qdisc_lookup(dev, autohandle))
749 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n,
752 const struct Qdisc_class_ops *cops;
757 if (n == 0 && len == 0)
759 drops = max_t(int, n, 0);
761 while ((parentid = sch->parent)) {
762 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
765 if (sch->flags & TCQ_F_NOPARENT)
767 /* TODO: perform the search on a per txq basis */
768 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
770 WARN_ON_ONCE(parentid != TC_H_ROOT);
773 cops = sch->ops->cl_ops;
774 if (cops->qlen_notify) {
775 cl = cops->get(sch, parentid);
776 cops->qlen_notify(sch, cl);
780 sch->qstats.backlog -= len;
781 __qdisc_qstats_drop(sch, drops);
785 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
787 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
788 struct nlmsghdr *n, u32 clid,
789 struct Qdisc *old, struct Qdisc *new)
792 qdisc_notify(net, skb, n, clid, old, new);
798 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
801 * When appropriate send a netlink notification using 'skb'
804 * On success, destroy old qdisc.
807 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
808 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
809 struct Qdisc *new, struct Qdisc *old)
811 struct Qdisc *q = old;
812 struct net *net = dev_net(dev);
815 if (parent == NULL) {
816 unsigned int i, num_q, ingress;
819 num_q = dev->num_tx_queues;
820 if ((q && q->flags & TCQ_F_INGRESS) ||
821 (new && new->flags & TCQ_F_INGRESS)) {
824 if (!dev_ingress_queue(dev))
828 if (dev->flags & IFF_UP)
831 if (new && new->ops->attach)
834 for (i = 0; i < num_q; i++) {
835 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
838 dev_queue = netdev_get_tx_queue(dev, i);
840 old = dev_graft_qdisc(dev_queue, new);
842 atomic_inc(&new->refcnt);
850 notify_and_destroy(net, skb, n, classid,
852 if (new && !new->ops->attach)
853 atomic_inc(&new->refcnt);
854 dev->qdisc = new ? : &noop_qdisc;
856 if (new && new->ops->attach)
857 new->ops->attach(new);
859 notify_and_destroy(net, skb, n, classid, old, new);
862 if (dev->flags & IFF_UP)
865 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
868 if (cops && cops->graft) {
869 unsigned long cl = cops->get(parent, classid);
871 err = cops->graft(parent, cl, new, &old);
872 cops->put(parent, cl);
877 notify_and_destroy(net, skb, n, classid, old, new);
882 /* lockdep annotation is needed for ingress; egress gets it only for name */
883 static struct lock_class_key qdisc_tx_lock;
884 static struct lock_class_key qdisc_rx_lock;
887 Allocate and initialize new qdisc.
889 Parameters are passed via opt.
892 static struct Qdisc *qdisc_create(struct net_device *dev,
893 struct netdev_queue *dev_queue,
894 struct Qdisc *p, u32 parent, u32 handle,
895 struct nlattr **tca, int *errp)
898 struct nlattr *kind = tca[TCA_KIND];
900 struct Qdisc_ops *ops;
901 struct qdisc_size_table *stab;
903 ops = qdisc_lookup_ops(kind);
904 #ifdef CONFIG_MODULES
905 if (ops == NULL && kind != NULL) {
907 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
908 /* We dropped the RTNL semaphore in order to
909 * perform the module load. So, even if we
910 * succeeded in loading the module we have to
911 * tell the caller to replay the request. We
912 * indicate this using -EAGAIN.
913 * We replay the request because the device may
914 * go away in the mean time.
917 request_module("sch_%s", name);
919 ops = qdisc_lookup_ops(kind);
921 /* We will try again qdisc_lookup_ops,
922 * so don't keep a reference.
924 module_put(ops->owner);
936 sch = qdisc_alloc(dev_queue, ops);
942 sch->parent = parent;
944 if (handle == TC_H_INGRESS) {
945 sch->flags |= TCQ_F_INGRESS;
946 handle = TC_H_MAKE(TC_H_INGRESS, 0);
947 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
950 handle = qdisc_alloc_handle(dev);
955 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
956 if (!netif_is_multiqueue(dev))
957 sch->flags |= TCQ_F_ONETXQUEUE;
960 sch->handle = handle;
962 /* This exist to keep backward compatible with a userspace
963 * loophole, what allowed userspace to get IFF_NO_QUEUE
964 * facility on older kernels by setting tx_queue_len=0 (prior
965 * to qdisc init), and then forgot to reinit tx_queue_len
966 * before again attaching a qdisc.
968 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
969 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
970 netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
973 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
974 if (qdisc_is_percpu_stats(sch)) {
976 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
977 if (!sch->cpu_bstats)
980 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
981 if (!sch->cpu_qstats)
986 stab = qdisc_get_stab(tca[TCA_STAB]);
991 rcu_assign_pointer(sch->stab, stab);
997 if (sch->flags & TCQ_F_MQROOT)
1000 if ((sch->parent != TC_H_ROOT) &&
1001 !(sch->flags & TCQ_F_INGRESS) &&
1002 (!p || !(p->flags & TCQ_F_MQROOT)))
1003 running = qdisc_root_sleeping_running(sch);
1005 running = &sch->running;
1007 err = gen_new_estimator(&sch->bstats,
1017 qdisc_hash_add(sch, false);
1021 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1025 kfree((char *) sch - sch->padded);
1027 module_put(ops->owner);
1033 free_percpu(sch->cpu_bstats);
1034 free_percpu(sch->cpu_qstats);
1036 * Any broken qdiscs that would require a ops->reset() here?
1037 * The qdisc was never in action so it shouldn't be necessary.
1039 qdisc_put_stab(rtnl_dereference(sch->stab));
1045 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
1047 struct qdisc_size_table *ostab, *stab = NULL;
1050 if (tca[TCA_OPTIONS]) {
1051 if (sch->ops->change == NULL)
1053 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1058 if (tca[TCA_STAB]) {
1059 stab = qdisc_get_stab(tca[TCA_STAB]);
1061 return PTR_ERR(stab);
1064 ostab = rtnl_dereference(sch->stab);
1065 rcu_assign_pointer(sch->stab, stab);
1066 qdisc_put_stab(ostab);
1068 if (tca[TCA_RATE]) {
1069 /* NB: ignores errors from replace_estimator
1070 because change can't be undone. */
1071 if (sch->flags & TCQ_F_MQROOT)
1073 gen_replace_estimator(&sch->bstats,
1077 qdisc_root_sleeping_running(sch),
1084 struct check_loop_arg {
1085 struct qdisc_walker w;
1090 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1091 struct qdisc_walker *w);
1093 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1095 struct check_loop_arg arg;
1097 if (q->ops->cl_ops == NULL)
1100 arg.w.stop = arg.w.skip = arg.w.count = 0;
1101 arg.w.fn = check_loop_fn;
1104 q->ops->cl_ops->walk(q, &arg.w);
1105 return arg.w.stop ? -ELOOP : 0;
1109 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1112 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1113 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1115 leaf = cops->leaf(q, cl);
1117 if (leaf == arg->p || arg->depth > 7)
1119 return check_loop(leaf, arg->p, arg->depth + 1);
1128 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1130 struct net *net = sock_net(skb->sk);
1131 struct tcmsg *tcm = nlmsg_data(n);
1132 struct nlattr *tca[TCA_MAX + 1];
1133 struct net_device *dev;
1135 struct Qdisc *q = NULL;
1136 struct Qdisc *p = NULL;
1139 if ((n->nlmsg_type != RTM_GETQDISC) &&
1140 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1143 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1147 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1151 clid = tcm->tcm_parent;
1153 if (clid != TC_H_ROOT) {
1154 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1155 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1158 q = qdisc_leaf(p, clid);
1159 } else if (dev_ingress_queue(dev)) {
1160 q = dev_ingress_queue(dev)->qdisc_sleeping;
1168 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1171 q = qdisc_lookup(dev, tcm->tcm_handle);
1176 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1179 if (n->nlmsg_type == RTM_DELQDISC) {
1184 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1188 qdisc_notify(net, skb, n, clid, NULL, q);
1194 * Create/change qdisc.
1197 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1199 struct net *net = sock_net(skb->sk);
1201 struct nlattr *tca[TCA_MAX + 1];
1202 struct net_device *dev;
1204 struct Qdisc *q, *p;
1207 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1211 /* Reinit, just in case something touches this. */
1212 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1216 tcm = nlmsg_data(n);
1217 clid = tcm->tcm_parent;
1220 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1226 if (clid != TC_H_ROOT) {
1227 if (clid != TC_H_INGRESS) {
1228 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1231 q = qdisc_leaf(p, clid);
1232 } else if (dev_ingress_queue_create(dev)) {
1233 q = dev_ingress_queue(dev)->qdisc_sleeping;
1239 /* It may be default qdisc, ignore it */
1240 if (q && q->handle == 0)
1243 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1244 if (tcm->tcm_handle) {
1245 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1247 if (TC_H_MIN(tcm->tcm_handle))
1249 q = qdisc_lookup(dev, tcm->tcm_handle);
1251 goto create_n_graft;
1252 if (n->nlmsg_flags & NLM_F_EXCL)
1254 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1257 (p && check_loop(q, p, 0)))
1259 atomic_inc(&q->refcnt);
1263 goto create_n_graft;
1265 /* This magic test requires explanation.
1267 * We know, that some child q is already
1268 * attached to this parent and have choice:
1269 * either to change it or to create/graft new one.
1271 * 1. We are allowed to create/graft only
1272 * if CREATE and REPLACE flags are set.
1274 * 2. If EXCL is set, requestor wanted to say,
1275 * that qdisc tcm_handle is not expected
1276 * to exist, so that we choose create/graft too.
1278 * 3. The last case is when no flags are set.
1279 * Alas, it is sort of hole in API, we
1280 * cannot decide what to do unambiguously.
1281 * For now we select create/graft, if
1282 * user gave KIND, which does not match existing.
1284 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1285 (n->nlmsg_flags & NLM_F_REPLACE) &&
1286 ((n->nlmsg_flags & NLM_F_EXCL) ||
1288 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1289 goto create_n_graft;
1293 if (!tcm->tcm_handle)
1295 q = qdisc_lookup(dev, tcm->tcm_handle);
1298 /* Change qdisc parameters */
1301 if (n->nlmsg_flags & NLM_F_EXCL)
1303 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1305 err = qdisc_change(q, tca);
1307 qdisc_notify(net, skb, n, clid, NULL, q);
1311 if (!(n->nlmsg_flags & NLM_F_CREATE))
1313 if (clid == TC_H_INGRESS) {
1314 if (dev_ingress_queue(dev))
1315 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1316 tcm->tcm_parent, tcm->tcm_parent,
1321 struct netdev_queue *dev_queue;
1323 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1324 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1326 dev_queue = p->dev_queue;
1328 dev_queue = netdev_get_tx_queue(dev, 0);
1330 q = qdisc_create(dev, dev_queue, p,
1331 tcm->tcm_parent, tcm->tcm_handle,
1341 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1351 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1352 u32 portid, u32 seq, u16 flags, int event)
1354 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
1355 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
1357 struct nlmsghdr *nlh;
1358 unsigned char *b = skb_tail_pointer(skb);
1360 struct qdisc_size_table *stab;
1364 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1366 goto out_nlmsg_trim;
1367 tcm = nlmsg_data(nlh);
1368 tcm->tcm_family = AF_UNSPEC;
1371 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1372 tcm->tcm_parent = clid;
1373 tcm->tcm_handle = q->handle;
1374 tcm->tcm_info = atomic_read(&q->refcnt);
1375 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1376 goto nla_put_failure;
1377 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1378 goto nla_put_failure;
1381 stab = rtnl_dereference(q->stab);
1382 if (stab && qdisc_dump_stab(skb, stab) < 0)
1383 goto nla_put_failure;
1385 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1386 NULL, &d, TCA_PAD) < 0)
1387 goto nla_put_failure;
1389 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1390 goto nla_put_failure;
1392 if (qdisc_is_percpu_stats(q)) {
1393 cpu_bstats = q->cpu_bstats;
1394 cpu_qstats = q->cpu_qstats;
1397 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
1398 &d, cpu_bstats, &q->bstats) < 0 ||
1399 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
1400 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
1401 goto nla_put_failure;
1403 if (gnet_stats_finish_copy(&d) < 0)
1404 goto nla_put_failure;
1406 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1415 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
1417 if (q->flags & TCQ_F_BUILTIN)
1419 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
1425 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1426 struct nlmsghdr *n, u32 clid,
1427 struct Qdisc *old, struct Qdisc *new)
1429 struct sk_buff *skb;
1430 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1432 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1436 if (old && !tc_qdisc_dump_ignore(old, false)) {
1437 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1438 0, RTM_DELQDISC) < 0)
1441 if (new && !tc_qdisc_dump_ignore(new, false)) {
1442 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1443 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1448 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1449 n->nlmsg_flags & NLM_F_ECHO);
1456 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1457 struct netlink_callback *cb,
1458 int *q_idx_p, int s_q_idx, bool recur,
1459 bool dump_invisible)
1461 int ret = 0, q_idx = *q_idx_p;
1469 if (q_idx < s_q_idx) {
1472 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1473 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1474 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1480 /* If dumping singletons, there is no qdisc_dev(root) and the singleton
1481 * itself has already been dumped.
1483 * If we've already dumped the top-level (ingress) qdisc above and the global
1484 * qdisc hashtable, we don't want to hit it again
1486 if (!qdisc_dev(root) || !recur)
1489 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1490 if (q_idx < s_q_idx) {
1494 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1495 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1496 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1510 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1512 struct net *net = sock_net(skb->sk);
1515 struct net_device *dev;
1516 const struct nlmsghdr *nlh = cb->nlh;
1517 struct tcmsg *tcm = nlmsg_data(nlh);
1518 struct nlattr *tca[TCA_MAX + 1];
1521 s_idx = cb->args[0];
1522 s_q_idx = q_idx = cb->args[1];
1527 err = nlmsg_parse(nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1531 for_each_netdev(net, dev) {
1532 struct netdev_queue *dev_queue;
1540 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1541 true, tca[TCA_DUMP_INVISIBLE]) < 0)
1544 dev_queue = dev_ingress_queue(dev);
1546 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1547 &q_idx, s_q_idx, false,
1548 tca[TCA_DUMP_INVISIBLE]) < 0)
1557 cb->args[1] = q_idx;
1564 /************************************************
1565 * Traffic classes manipulation. *
1566 ************************************************/
1570 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1572 struct net *net = sock_net(skb->sk);
1573 struct tcmsg *tcm = nlmsg_data(n);
1574 struct nlattr *tca[TCA_MAX + 1];
1575 struct net_device *dev;
1576 struct Qdisc *q = NULL;
1577 const struct Qdisc_class_ops *cops;
1578 unsigned long cl = 0;
1579 unsigned long new_cl;
1585 if ((n->nlmsg_type != RTM_GETTCLASS) &&
1586 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1589 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1593 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1598 parent == TC_H_UNSPEC - unspecified parent.
1599 parent == TC_H_ROOT - class is root, which has no parent.
1600 parent == X:0 - parent is root class.
1601 parent == X:Y - parent is a node in hierarchy.
1602 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1604 handle == 0:0 - generate handle from kernel pool.
1605 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1606 handle == X:Y - clear.
1607 handle == X:0 - root class.
1610 /* Step 1. Determine qdisc handle X:0 */
1612 portid = tcm->tcm_parent;
1613 clid = tcm->tcm_handle;
1614 qid = TC_H_MAJ(clid);
1616 if (portid != TC_H_ROOT) {
1617 u32 qid1 = TC_H_MAJ(portid);
1620 /* If both majors are known, they must be identical. */
1625 } else if (qid == 0)
1626 qid = dev->qdisc->handle;
1628 /* Now qid is genuine qdisc handle consistent
1629 * both with parent and child.
1631 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1634 portid = TC_H_MAKE(qid, portid);
1637 qid = dev->qdisc->handle;
1640 /* OK. Locate qdisc */
1641 q = qdisc_lookup(dev, qid);
1645 /* An check that it supports classes */
1646 cops = q->ops->cl_ops;
1650 /* Now try to get class */
1652 if (portid == TC_H_ROOT)
1655 clid = TC_H_MAKE(qid, clid);
1658 cl = cops->get(q, clid);
1662 if (n->nlmsg_type != RTM_NEWTCLASS ||
1663 !(n->nlmsg_flags & NLM_F_CREATE))
1666 switch (n->nlmsg_type) {
1669 if (n->nlmsg_flags & NLM_F_EXCL)
1675 err = cops->delete(q, cl);
1677 tclass_notify(net, skb, n, q, cl,
1681 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1692 err = cops->change(q, clid, portid, tca, &new_cl);
1694 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1704 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1706 u32 portid, u32 seq, u16 flags, int event)
1709 struct nlmsghdr *nlh;
1710 unsigned char *b = skb_tail_pointer(skb);
1712 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1715 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1717 goto out_nlmsg_trim;
1718 tcm = nlmsg_data(nlh);
1719 tcm->tcm_family = AF_UNSPEC;
1722 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1723 tcm->tcm_parent = q->handle;
1724 tcm->tcm_handle = q->handle;
1726 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1727 goto nla_put_failure;
1728 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1729 goto nla_put_failure;
1731 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1732 NULL, &d, TCA_PAD) < 0)
1733 goto nla_put_failure;
1735 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1736 goto nla_put_failure;
1738 if (gnet_stats_finish_copy(&d) < 0)
1739 goto nla_put_failure;
1741 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1750 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1751 struct nlmsghdr *n, struct Qdisc *q,
1752 unsigned long cl, int event)
1754 struct sk_buff *skb;
1755 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1757 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1761 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1766 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1767 n->nlmsg_flags & NLM_F_ECHO);
1770 struct qdisc_dump_args {
1771 struct qdisc_walker w;
1772 struct sk_buff *skb;
1773 struct netlink_callback *cb;
1776 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
1777 struct qdisc_walker *arg)
1779 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1781 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1782 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1786 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1787 struct tcmsg *tcm, struct netlink_callback *cb,
1790 struct qdisc_dump_args arg;
1792 if (tc_qdisc_dump_ignore(q, false) ||
1793 *t_p < s_t || !q->ops->cl_ops ||
1795 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1800 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1801 arg.w.fn = qdisc_class_dump;
1805 arg.w.skip = cb->args[1];
1807 q->ops->cl_ops->walk(q, &arg.w);
1808 cb->args[1] = arg.w.count;
1815 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1816 struct tcmsg *tcm, struct netlink_callback *cb,
1825 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1828 if (!qdisc_dev(root))
1831 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1832 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1839 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1841 struct tcmsg *tcm = nlmsg_data(cb->nlh);
1842 struct net *net = sock_net(skb->sk);
1843 struct netdev_queue *dev_queue;
1844 struct net_device *dev;
1847 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1849 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1856 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1859 dev_queue = dev_ingress_queue(dev);
1861 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1872 /* Main classifier routine: scans classifier chain attached
1873 * to this qdisc, (optionally) tests for protocol and asks
1874 * specific classifiers.
1876 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1877 struct tcf_result *res, bool compat_mode)
1879 __be16 protocol = tc_skb_protocol(skb);
1880 #ifdef CONFIG_NET_CLS_ACT
1881 const int max_reclassify_loop = 4;
1882 const struct tcf_proto *old_tp = tp;
1887 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1890 if (tp->protocol != protocol &&
1891 tp->protocol != htons(ETH_P_ALL))
1894 err = tp->classify(skb, tp, res);
1895 #ifdef CONFIG_NET_CLS_ACT
1896 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode))
1903 return TC_ACT_UNSPEC; /* signal: continue lookup */
1904 #ifdef CONFIG_NET_CLS_ACT
1906 if (unlikely(limit++ >= max_reclassify_loop)) {
1907 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
1908 tp->q->ops->id, tp->prio & 0xffff,
1909 ntohs(tp->protocol));
1914 protocol = tc_skb_protocol(skb);
1918 EXPORT_SYMBOL(tc_classify);
1920 #ifdef CONFIG_PROC_FS
1921 static int psched_show(struct seq_file *seq, void *v)
1923 seq_printf(seq, "%08x %08x %08x %08x\n",
1924 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1926 (u32)NSEC_PER_SEC / hrtimer_resolution);
1931 static int psched_open(struct inode *inode, struct file *file)
1933 return single_open(file, psched_show, NULL);
1936 static const struct file_operations psched_fops = {
1937 .owner = THIS_MODULE,
1938 .open = psched_open,
1940 .llseek = seq_lseek,
1941 .release = single_release,
1944 static int __net_init psched_net_init(struct net *net)
1946 struct proc_dir_entry *e;
1948 e = proc_create("psched", 0, net->proc_net, &psched_fops);
1955 static void __net_exit psched_net_exit(struct net *net)
1957 remove_proc_entry("psched", net->proc_net);
1960 static int __net_init psched_net_init(struct net *net)
1965 static void __net_exit psched_net_exit(struct net *net)
1970 static struct pernet_operations psched_net_ops = {
1971 .init = psched_net_init,
1972 .exit = psched_net_exit,
1975 static int __init pktsched_init(void)
1979 err = register_pernet_subsys(&psched_net_ops);
1981 pr_err("pktsched_init: "
1982 "cannot initialize per netns operations\n");
1986 register_qdisc(&pfifo_fast_ops);
1987 register_qdisc(&pfifo_qdisc_ops);
1988 register_qdisc(&bfifo_qdisc_ops);
1989 register_qdisc(&pfifo_head_drop_qdisc_ops);
1990 register_qdisc(&mq_qdisc_ops);
1991 register_qdisc(&noqueue_qdisc_ops);
1993 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1994 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1995 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
1997 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1998 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1999 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2005 subsys_initcall(pktsched_init);