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>
31 #include <net/net_namespace.h>
33 #include <net/netlink.h>
34 #include <net/pkt_sched.h>
36 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid,
37 struct Qdisc *old, struct Qdisc *new);
38 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
39 struct Qdisc *q, unsigned long cl, int event);
46 This file consists of two interrelated parts:
48 1. queueing disciplines manager frontend.
49 2. traffic classes manager frontend.
51 Generally, queueing discipline ("qdisc") is a black box,
52 which is able to enqueue packets and to dequeue them (when
53 device is ready to send something) in order and at times
54 determined by algorithm hidden in it.
56 qdisc's are divided to two categories:
57 - "queues", which have no internal structure visible from outside.
58 - "schedulers", which split all the packets to "traffic classes",
59 using "packet classifiers" (look at cls_api.c)
61 In turn, classes may have child qdiscs (as rule, queues)
62 attached to them etc. etc. etc.
64 The goal of the routines in this file is to translate
65 information supplied by user in the form of handles
66 to more intelligible for kernel form, to make some sanity
67 checks and part of work, which is common to all qdiscs
68 and to provide rtnetlink notifications.
70 All real intelligent work is done inside qdisc modules.
74 Every discipline has two major routines: enqueue and dequeue.
78 dequeue usually returns a skb to send. It is allowed to return NULL,
79 but it does not mean that queue is empty, it just means that
80 discipline does not want to send anything this time.
81 Queue is really empty if q->q.qlen == 0.
82 For complicated disciplines with multiple queues q->q is not
83 real packet queue, but however q->q.qlen must be valid.
87 enqueue returns 0, if packet was enqueued successfully.
88 If packet (this one or another one) was dropped, it returns
90 NET_XMIT_DROP - this packet dropped
91 Expected action: do not backoff, but wait until queue will clear.
92 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
93 Expected action: backoff or ignore
94 NET_XMIT_POLICED - dropped by police.
95 Expected action: backoff or error to real-time apps.
101 requeues once dequeued packet. It is used for non-standard or
102 just buggy devices, which can defer output even if netif_queue_stopped()=0.
106 returns qdisc to initial state: purge all buffers, clear all
107 timers, counters (except for statistics) etc.
111 initializes newly created qdisc.
115 destroys resources allocated by init and during lifetime of qdisc.
119 changes qdisc parameters.
122 /* Protects list of registered TC modules. It is pure SMP lock. */
123 static DEFINE_RWLOCK(qdisc_mod_lock);
126 /************************************************
127 * Queueing disciplines manipulation. *
128 ************************************************/
131 /* The list of all installed queueing disciplines. */
133 static struct Qdisc_ops *qdisc_base;
135 /* Register/uregister queueing discipline */
137 int register_qdisc(struct Qdisc_ops *qops)
139 struct Qdisc_ops *q, **qp;
142 write_lock(&qdisc_mod_lock);
143 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
144 if (!strcmp(qops->id, q->id))
147 if (qops->enqueue == NULL)
148 qops->enqueue = noop_qdisc_ops.enqueue;
149 if (qops->requeue == NULL)
150 qops->requeue = noop_qdisc_ops.requeue;
151 if (qops->dequeue == NULL)
152 qops->dequeue = noop_qdisc_ops.dequeue;
158 write_unlock(&qdisc_mod_lock);
161 EXPORT_SYMBOL(register_qdisc);
163 int unregister_qdisc(struct Qdisc_ops *qops)
165 struct Qdisc_ops *q, **qp;
168 write_lock(&qdisc_mod_lock);
169 for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next)
177 write_unlock(&qdisc_mod_lock);
180 EXPORT_SYMBOL(unregister_qdisc);
182 /* We know handle. Find qdisc among all qdisc's attached to device
183 (root qdisc, all its children, children of children etc.)
186 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
190 list_for_each_entry(q, &dev->qdisc_list, list) {
191 if (q->handle == handle)
197 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
201 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
205 cl = cops->get(p, classid);
209 leaf = cops->leaf(p, cl);
214 /* Find queueing discipline by name */
216 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
218 struct Qdisc_ops *q = NULL;
221 read_lock(&qdisc_mod_lock);
222 for (q = qdisc_base; q; q = q->next) {
223 if (nla_strcmp(kind, q->id) == 0) {
224 if (!try_module_get(q->owner))
229 read_unlock(&qdisc_mod_lock);
234 static struct qdisc_rate_table *qdisc_rtab_list;
236 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
238 struct qdisc_rate_table *rtab;
240 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
241 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
247 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
248 nla_len(tab) != TC_RTAB_SIZE)
251 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
255 memcpy(rtab->data, nla_data(tab), 1024);
256 rtab->next = qdisc_rtab_list;
257 qdisc_rtab_list = rtab;
261 EXPORT_SYMBOL(qdisc_get_rtab);
263 void qdisc_put_rtab(struct qdisc_rate_table *tab)
265 struct qdisc_rate_table *rtab, **rtabp;
267 if (!tab || --tab->refcnt)
270 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
278 EXPORT_SYMBOL(qdisc_put_rtab);
280 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
282 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
285 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
287 __netif_schedule(wd->qdisc);
289 return HRTIMER_NORESTART;
292 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
294 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
295 wd->timer.function = qdisc_watchdog;
298 EXPORT_SYMBOL(qdisc_watchdog_init);
300 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
304 wd->qdisc->flags |= TCQ_F_THROTTLED;
305 time = ktime_set(0, 0);
306 time = ktime_add_ns(time, PSCHED_US2NS(expires));
307 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
309 EXPORT_SYMBOL(qdisc_watchdog_schedule);
311 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
313 hrtimer_cancel(&wd->timer);
314 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
316 EXPORT_SYMBOL(qdisc_watchdog_cancel);
318 struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
320 unsigned int size = n * sizeof(struct hlist_head), i;
321 struct hlist_head *h;
323 if (size <= PAGE_SIZE)
324 h = kmalloc(size, GFP_KERNEL);
326 h = (struct hlist_head *)
327 __get_free_pages(GFP_KERNEL, get_order(size));
330 for (i = 0; i < n; i++)
331 INIT_HLIST_HEAD(&h[i]);
336 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
338 unsigned int size = n * sizeof(struct hlist_head);
340 if (size <= PAGE_SIZE)
343 free_pages((unsigned long)h, get_order(size));
346 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
348 struct Qdisc_class_common *cl;
349 struct hlist_node *n, *next;
350 struct hlist_head *nhash, *ohash;
351 unsigned int nsize, nmask, osize;
354 /* Rehash when load factor exceeds 0.75 */
355 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
357 nsize = clhash->hashsize * 2;
359 nhash = qdisc_class_hash_alloc(nsize);
363 ohash = clhash->hash;
364 osize = clhash->hashsize;
367 for (i = 0; i < osize; i++) {
368 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
369 h = qdisc_class_hash(cl->classid, nmask);
370 hlist_add_head(&cl->hnode, &nhash[h]);
373 clhash->hash = nhash;
374 clhash->hashsize = nsize;
375 clhash->hashmask = nmask;
376 sch_tree_unlock(sch);
378 qdisc_class_hash_free(ohash, osize);
380 EXPORT_SYMBOL(qdisc_class_hash_grow);
382 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
384 unsigned int size = 4;
386 clhash->hash = qdisc_class_hash_alloc(size);
387 if (clhash->hash == NULL)
389 clhash->hashsize = size;
390 clhash->hashmask = size - 1;
391 clhash->hashelems = 0;
394 EXPORT_SYMBOL(qdisc_class_hash_init);
396 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
398 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
400 EXPORT_SYMBOL(qdisc_class_hash_destroy);
402 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
403 struct Qdisc_class_common *cl)
407 INIT_HLIST_NODE(&cl->hnode);
408 h = qdisc_class_hash(cl->classid, clhash->hashmask);
409 hlist_add_head(&cl->hnode, &clhash->hash[h]);
412 EXPORT_SYMBOL(qdisc_class_hash_insert);
414 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
415 struct Qdisc_class_common *cl)
417 hlist_del(&cl->hnode);
420 EXPORT_SYMBOL(qdisc_class_hash_remove);
422 /* Allocate an unique handle from space managed by kernel */
424 static u32 qdisc_alloc_handle(struct net_device *dev)
427 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
430 autohandle += TC_H_MAKE(0x10000U, 0);
431 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
432 autohandle = TC_H_MAKE(0x80000000U, 0);
433 } while (qdisc_lookup(dev, autohandle) && --i > 0);
435 return i>0 ? autohandle : 0;
438 /* Attach toplevel qdisc to device dev */
440 static struct Qdisc *
441 dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc)
443 struct netdev_queue *dev_queue;
444 spinlock_t *root_lock;
445 struct Qdisc *oqdisc;
448 if (dev->flags & IFF_UP)
452 if (qdisc && qdisc->flags&TCQ_F_INGRESS)
456 dev_queue = &dev->rx_queue;
457 oqdisc = dev_queue->qdisc;
459 dev_queue = netdev_get_tx_queue(dev, 0);
460 oqdisc = dev_queue->qdisc_sleeping;
463 root_lock = qdisc_root_lock(oqdisc);
464 spin_lock_bh(root_lock);
467 /* Prune old scheduler */
468 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) {
471 dev_queue->qdisc = NULL;
473 dev_queue->qdisc = qdisc;
477 /* Prune old scheduler */
478 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
481 /* ... and graft new one */
484 dev_queue->qdisc_sleeping = qdisc;
485 dev_queue->qdisc = &noop_qdisc;
488 spin_unlock_bh(root_lock);
490 if (dev->flags & IFF_UP)
496 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
498 const struct Qdisc_class_ops *cops;
504 while ((parentid = sch->parent)) {
505 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
508 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
510 WARN_ON(parentid != TC_H_ROOT);
513 cops = sch->ops->cl_ops;
514 if (cops->qlen_notify) {
515 cl = cops->get(sch, parentid);
516 cops->qlen_notify(sch, cl);
522 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
524 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
527 Old qdisc is not destroyed but returned in *old.
530 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
532 struct Qdisc *new, struct Qdisc **old)
535 struct Qdisc *q = *old;
538 if (parent == NULL) {
539 if (q && q->flags&TCQ_F_INGRESS) {
540 *old = dev_graft_qdisc(dev, q);
542 *old = dev_graft_qdisc(dev, new);
545 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
550 unsigned long cl = cops->get(parent, classid);
552 err = cops->graft(parent, cl, new, old);
553 cops->put(parent, cl);
561 Allocate and initialize new qdisc.
563 Parameters are passed via opt.
566 static struct Qdisc *
567 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
568 u32 parent, u32 handle, struct nlattr **tca, int *errp)
571 struct nlattr *kind = tca[TCA_KIND];
573 struct Qdisc_ops *ops;
575 ops = qdisc_lookup_ops(kind);
577 if (ops == NULL && kind != NULL) {
579 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
580 /* We dropped the RTNL semaphore in order to
581 * perform the module load. So, even if we
582 * succeeded in loading the module we have to
583 * tell the caller to replay the request. We
584 * indicate this using -EAGAIN.
585 * We replay the request because the device may
586 * go away in the mean time.
589 request_module("sch_%s", name);
591 ops = qdisc_lookup_ops(kind);
593 /* We will try again qdisc_lookup_ops,
594 * so don't keep a reference.
596 module_put(ops->owner);
608 sch = qdisc_alloc(dev_queue, ops);
614 sch->parent = parent;
616 if (handle == TC_H_INGRESS) {
617 sch->flags |= TCQ_F_INGRESS;
618 handle = TC_H_MAKE(TC_H_INGRESS, 0);
621 handle = qdisc_alloc_handle(dev);
628 sch->handle = handle;
630 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
632 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
633 qdisc_root_lock(sch),
637 * Any broken qdiscs that would require
638 * a ops->reset() here? The qdisc was never
639 * in action so it shouldn't be necessary.
646 spin_lock_bh(&dev->qdisc_list_lock);
647 list_add_tail(&sch->list, &dev->qdisc_list);
648 spin_unlock_bh(&dev->qdisc_list_lock);
654 kfree((char *) sch - sch->padded);
656 module_put(ops->owner);
662 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
664 if (tca[TCA_OPTIONS]) {
667 if (sch->ops->change == NULL)
669 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
674 gen_replace_estimator(&sch->bstats, &sch->rate_est,
675 qdisc_root_lock(sch), tca[TCA_RATE]);
679 struct check_loop_arg
681 struct qdisc_walker w;
686 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
688 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
690 struct check_loop_arg arg;
692 if (q->ops->cl_ops == NULL)
695 arg.w.stop = arg.w.skip = arg.w.count = 0;
696 arg.w.fn = check_loop_fn;
699 q->ops->cl_ops->walk(q, &arg.w);
700 return arg.w.stop ? -ELOOP : 0;
704 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
707 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
708 struct check_loop_arg *arg = (struct check_loop_arg *)w;
710 leaf = cops->leaf(q, cl);
712 if (leaf == arg->p || arg->depth > 7)
714 return check_loop(leaf, arg->p, arg->depth + 1);
723 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
725 struct net *net = sock_net(skb->sk);
726 struct tcmsg *tcm = NLMSG_DATA(n);
727 struct nlattr *tca[TCA_MAX + 1];
728 struct net_device *dev;
729 u32 clid = tcm->tcm_parent;
730 struct Qdisc *q = NULL;
731 struct Qdisc *p = NULL;
734 if (net != &init_net)
737 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
740 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
745 if (clid != TC_H_ROOT) {
746 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
747 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
749 q = qdisc_leaf(p, clid);
750 } else { /* ingress */
751 q = dev->rx_queue.qdisc;
754 struct netdev_queue *dev_queue;
755 dev_queue = netdev_get_tx_queue(dev, 0);
756 q = dev_queue->qdisc_sleeping;
761 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
764 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
768 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
771 if (n->nlmsg_type == RTM_DELQDISC) {
776 if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0)
779 spinlock_t *root_lock = qdisc_root_lock(q);
781 qdisc_notify(skb, n, clid, q, NULL);
782 spin_unlock_bh(root_lock);
784 spin_unlock_bh(root_lock);
787 qdisc_notify(skb, n, clid, NULL, q);
796 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
798 struct net *net = sock_net(skb->sk);
800 struct nlattr *tca[TCA_MAX + 1];
801 struct net_device *dev;
806 if (net != &init_net)
810 /* Reinit, just in case something touches this. */
812 clid = tcm->tcm_parent;
815 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
818 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
823 if (clid != TC_H_ROOT) {
824 if (clid != TC_H_INGRESS) {
825 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
827 q = qdisc_leaf(p, clid);
828 } else { /*ingress */
829 q = dev->rx_queue.qdisc;
832 struct netdev_queue *dev_queue;
833 dev_queue = netdev_get_tx_queue(dev, 0);
834 q = dev_queue->qdisc_sleeping;
837 /* It may be default qdisc, ignore it */
838 if (q && q->handle == 0)
841 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
842 if (tcm->tcm_handle) {
843 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
845 if (TC_H_MIN(tcm->tcm_handle))
847 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
849 if (n->nlmsg_flags&NLM_F_EXCL)
851 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
854 (p && check_loop(q, p, 0)))
856 atomic_inc(&q->refcnt);
862 /* This magic test requires explanation.
864 * We know, that some child q is already
865 * attached to this parent and have choice:
866 * either to change it or to create/graft new one.
868 * 1. We are allowed to create/graft only
869 * if CREATE and REPLACE flags are set.
871 * 2. If EXCL is set, requestor wanted to say,
872 * that qdisc tcm_handle is not expected
873 * to exist, so that we choose create/graft too.
875 * 3. The last case is when no flags are set.
876 * Alas, it is sort of hole in API, we
877 * cannot decide what to do unambiguously.
878 * For now we select create/graft, if
879 * user gave KIND, which does not match existing.
881 if ((n->nlmsg_flags&NLM_F_CREATE) &&
882 (n->nlmsg_flags&NLM_F_REPLACE) &&
883 ((n->nlmsg_flags&NLM_F_EXCL) ||
885 nla_strcmp(tca[TCA_KIND], q->ops->id))))
890 if (!tcm->tcm_handle)
892 q = qdisc_lookup(dev, tcm->tcm_handle);
895 /* Change qdisc parameters */
898 if (n->nlmsg_flags&NLM_F_EXCL)
900 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
902 err = qdisc_change(q, tca);
904 qdisc_notify(skb, n, clid, NULL, q);
908 if (!(n->nlmsg_flags&NLM_F_CREATE))
910 if (clid == TC_H_INGRESS)
911 q = qdisc_create(dev, &dev->rx_queue,
912 tcm->tcm_parent, tcm->tcm_parent,
915 q = qdisc_create(dev, netdev_get_tx_queue(dev, 0),
916 tcm->tcm_parent, tcm->tcm_handle,
926 struct Qdisc *old_q = NULL;
927 spinlock_t *root_lock;
929 err = qdisc_graft(dev, p, clid, q, &old_q);
932 root_lock = qdisc_root_lock(q);
933 spin_lock_bh(root_lock);
935 spin_unlock_bh(root_lock);
939 qdisc_notify(skb, n, clid, old_q, q);
941 root_lock = qdisc_root_lock(old_q);
942 spin_lock_bh(root_lock);
943 qdisc_destroy(old_q);
944 spin_unlock_bh(root_lock);
950 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
951 u32 pid, u32 seq, u16 flags, int event)
954 struct nlmsghdr *nlh;
955 unsigned char *b = skb_tail_pointer(skb);
958 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
959 tcm = NLMSG_DATA(nlh);
960 tcm->tcm_family = AF_UNSPEC;
963 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
964 tcm->tcm_parent = clid;
965 tcm->tcm_handle = q->handle;
966 tcm->tcm_info = atomic_read(&q->refcnt);
967 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
968 if (q->ops->dump && q->ops->dump(q, skb) < 0)
969 goto nla_put_failure;
970 q->qstats.qlen = q->q.qlen;
972 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
973 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
974 goto nla_put_failure;
976 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
977 goto nla_put_failure;
979 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
980 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
981 gnet_stats_copy_queue(&d, &q->qstats) < 0)
982 goto nla_put_failure;
984 if (gnet_stats_finish_copy(&d) < 0)
985 goto nla_put_failure;
987 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
996 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
997 u32 clid, struct Qdisc *old, struct Qdisc *new)
1000 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1002 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1006 if (old && old->handle) {
1007 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
1011 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1016 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1023 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1025 struct net *net = sock_net(skb->sk);
1028 struct net_device *dev;
1031 if (net != &init_net)
1034 s_idx = cb->args[0];
1035 s_q_idx = q_idx = cb->args[1];
1036 read_lock(&dev_base_lock);
1038 for_each_netdev(&init_net, dev) {
1044 list_for_each_entry(q, &dev->qdisc_list, list) {
1045 if (q_idx < s_q_idx) {
1049 if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1050 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1059 read_unlock(&dev_base_lock);
1062 cb->args[1] = q_idx;
1069 /************************************************
1070 * Traffic classes manipulation. *
1071 ************************************************/
1075 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1077 struct net *net = sock_net(skb->sk);
1078 struct netdev_queue *dev_queue;
1079 struct tcmsg *tcm = NLMSG_DATA(n);
1080 struct nlattr *tca[TCA_MAX + 1];
1081 struct net_device *dev;
1082 struct Qdisc *q = NULL;
1083 const struct Qdisc_class_ops *cops;
1084 unsigned long cl = 0;
1085 unsigned long new_cl;
1086 u32 pid = tcm->tcm_parent;
1087 u32 clid = tcm->tcm_handle;
1088 u32 qid = TC_H_MAJ(clid);
1091 if (net != &init_net)
1094 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1097 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1102 parent == TC_H_UNSPEC - unspecified parent.
1103 parent == TC_H_ROOT - class is root, which has no parent.
1104 parent == X:0 - parent is root class.
1105 parent == X:Y - parent is a node in hierarchy.
1106 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1108 handle == 0:0 - generate handle from kernel pool.
1109 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1110 handle == X:Y - clear.
1111 handle == X:0 - root class.
1114 /* Step 1. Determine qdisc handle X:0 */
1116 dev_queue = netdev_get_tx_queue(dev, 0);
1117 if (pid != TC_H_ROOT) {
1118 u32 qid1 = TC_H_MAJ(pid);
1121 /* If both majors are known, they must be identical. */
1126 } else if (qid == 0)
1127 qid = dev_queue->qdisc_sleeping->handle;
1129 /* Now qid is genuine qdisc handle consistent
1130 both with parent and child.
1132 TC_H_MAJ(pid) still may be unspecified, complete it now.
1135 pid = TC_H_MAKE(qid, pid);
1138 qid = dev_queue->qdisc_sleeping->handle;
1141 /* OK. Locate qdisc */
1142 if ((q = qdisc_lookup(dev, qid)) == NULL)
1145 /* An check that it supports classes */
1146 cops = q->ops->cl_ops;
1150 /* Now try to get class */
1152 if (pid == TC_H_ROOT)
1155 clid = TC_H_MAKE(qid, clid);
1158 cl = cops->get(q, clid);
1162 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1165 switch (n->nlmsg_type) {
1168 if (n->nlmsg_flags&NLM_F_EXCL)
1172 err = cops->delete(q, cl);
1174 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1177 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1186 err = cops->change(q, clid, pid, tca, &new_cl);
1188 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1198 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1200 u32 pid, u32 seq, u16 flags, int event)
1203 struct nlmsghdr *nlh;
1204 unsigned char *b = skb_tail_pointer(skb);
1206 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1208 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1209 tcm = NLMSG_DATA(nlh);
1210 tcm->tcm_family = AF_UNSPEC;
1211 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1212 tcm->tcm_parent = q->handle;
1213 tcm->tcm_handle = q->handle;
1215 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1216 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1217 goto nla_put_failure;
1219 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1220 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
1221 goto nla_put_failure;
1223 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1224 goto nla_put_failure;
1226 if (gnet_stats_finish_copy(&d) < 0)
1227 goto nla_put_failure;
1229 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1238 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1239 struct Qdisc *q, unsigned long cl, int event)
1241 struct sk_buff *skb;
1242 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1244 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1248 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1253 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1256 struct qdisc_dump_args
1258 struct qdisc_walker w;
1259 struct sk_buff *skb;
1260 struct netlink_callback *cb;
1263 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1265 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1267 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1268 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1271 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1273 struct net *net = sock_net(skb->sk);
1276 struct net_device *dev;
1278 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1279 struct qdisc_dump_args arg;
1281 if (net != &init_net)
1284 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1286 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1292 list_for_each_entry(q, &dev->qdisc_list, list) {
1293 if (t < s_t || !q->ops->cl_ops ||
1295 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1300 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1301 arg.w.fn = qdisc_class_dump;
1305 arg.w.skip = cb->args[1];
1307 q->ops->cl_ops->walk(q, &arg.w);
1308 cb->args[1] = arg.w.count;
1320 /* Main classifier routine: scans classifier chain attached
1321 to this qdisc, (optionally) tests for protocol and asks
1322 specific classifiers.
1324 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1325 struct tcf_result *res)
1327 __be16 protocol = skb->protocol;
1330 for (; tp; tp = tp->next) {
1331 if ((tp->protocol == protocol ||
1332 tp->protocol == htons(ETH_P_ALL)) &&
1333 (err = tp->classify(skb, tp, res)) >= 0) {
1334 #ifdef CONFIG_NET_CLS_ACT
1335 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1336 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1343 EXPORT_SYMBOL(tc_classify_compat);
1345 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1346 struct tcf_result *res)
1350 #ifdef CONFIG_NET_CLS_ACT
1351 struct tcf_proto *otp = tp;
1354 protocol = skb->protocol;
1356 err = tc_classify_compat(skb, tp, res);
1357 #ifdef CONFIG_NET_CLS_ACT
1358 if (err == TC_ACT_RECLASSIFY) {
1359 u32 verd = G_TC_VERD(skb->tc_verd);
1362 if (verd++ >= MAX_REC_LOOP) {
1363 printk("rule prio %u protocol %02x reclassify loop, "
1365 tp->prio&0xffff, ntohs(tp->protocol));
1368 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1374 EXPORT_SYMBOL(tc_classify);
1376 void tcf_destroy(struct tcf_proto *tp)
1378 tp->ops->destroy(tp);
1379 module_put(tp->ops->owner);
1383 void tcf_destroy_chain(struct tcf_proto **fl)
1385 struct tcf_proto *tp;
1387 while ((tp = *fl) != NULL) {
1392 EXPORT_SYMBOL(tcf_destroy_chain);
1394 #ifdef CONFIG_PROC_FS
1395 static int psched_show(struct seq_file *seq, void *v)
1399 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1400 seq_printf(seq, "%08x %08x %08x %08x\n",
1401 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1403 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1408 static int psched_open(struct inode *inode, struct file *file)
1410 return single_open(file, psched_show, PDE(inode)->data);
1413 static const struct file_operations psched_fops = {
1414 .owner = THIS_MODULE,
1415 .open = psched_open,
1417 .llseek = seq_lseek,
1418 .release = single_release,
1422 static int __init pktsched_init(void)
1424 register_qdisc(&pfifo_qdisc_ops);
1425 register_qdisc(&bfifo_qdisc_ops);
1426 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1428 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1429 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1430 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1431 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1432 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1433 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1438 subsys_initcall(pktsched_init);