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>
38 #include <net/pkt_cls.h>
45 This file consists of two interrelated parts:
47 1. queueing disciplines manager frontend.
48 2. traffic classes manager frontend.
50 Generally, queueing discipline ("qdisc") is a black box,
51 which is able to enqueue packets and to dequeue them (when
52 device is ready to send something) in order and at times
53 determined by algorithm hidden in it.
55 qdisc's are divided to two categories:
56 - "queues", which have no internal structure visible from outside.
57 - "schedulers", which split all the packets to "traffic classes",
58 using "packet classifiers" (look at cls_api.c)
60 In turn, classes may have child qdiscs (as rule, queues)
61 attached to them etc. etc. etc.
63 The goal of the routines in this file is to translate
64 information supplied by user in the form of handles
65 to more intelligible for kernel form, to make some sanity
66 checks and part of work, which is common to all qdiscs
67 and to provide rtnetlink notifications.
69 All real intelligent work is done inside qdisc modules.
73 Every discipline has two major routines: enqueue and dequeue.
77 dequeue usually returns a skb to send. It is allowed to return NULL,
78 but it does not mean that queue is empty, it just means that
79 discipline does not want to send anything this time.
80 Queue is really empty if q->q.qlen == 0.
81 For complicated disciplines with multiple queues q->q is not
82 real packet queue, but however q->q.qlen must be valid.
86 enqueue returns 0, if packet was enqueued successfully.
87 If packet (this one or another one) was dropped, it returns
89 NET_XMIT_DROP - this packet dropped
90 Expected action: do not backoff, but wait until queue will clear.
91 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
92 Expected action: backoff or ignore
98 like dequeue but without removing a packet from the queue
102 returns qdisc to initial state: purge all buffers, clear all
103 timers, counters (except for statistics) etc.
107 initializes newly created qdisc.
111 destroys resources allocated by init and during lifetime of qdisc.
115 changes qdisc parameters.
118 /* Protects list of registered TC modules. It is pure SMP lock. */
119 static DEFINE_RWLOCK(qdisc_mod_lock);
122 /************************************************
123 * Queueing disciplines manipulation. *
124 ************************************************/
127 /* The list of all installed queueing disciplines. */
129 static struct Qdisc_ops *qdisc_base;
131 /* Register/unregister queueing discipline */
133 int register_qdisc(struct Qdisc_ops *qops)
135 struct Qdisc_ops *q, **qp;
138 write_lock(&qdisc_mod_lock);
139 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
140 if (!strcmp(qops->id, q->id))
143 if (qops->enqueue == NULL)
144 qops->enqueue = noop_qdisc_ops.enqueue;
145 if (qops->peek == NULL) {
146 if (qops->dequeue == NULL)
147 qops->peek = noop_qdisc_ops.peek;
151 if (qops->dequeue == NULL)
152 qops->dequeue = noop_qdisc_ops.dequeue;
155 const struct Qdisc_class_ops *cops = qops->cl_ops;
157 if (!(cops->find && cops->walk && cops->leaf))
160 if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
168 write_unlock(&qdisc_mod_lock);
175 EXPORT_SYMBOL(register_qdisc);
177 int unregister_qdisc(struct Qdisc_ops *qops)
179 struct Qdisc_ops *q, **qp;
182 write_lock(&qdisc_mod_lock);
183 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
191 write_unlock(&qdisc_mod_lock);
194 EXPORT_SYMBOL(unregister_qdisc);
196 /* Get default qdisc if not otherwise specified */
197 void qdisc_get_default(char *name, size_t len)
199 read_lock(&qdisc_mod_lock);
200 strlcpy(name, default_qdisc_ops->id, len);
201 read_unlock(&qdisc_mod_lock);
204 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
206 struct Qdisc_ops *q = NULL;
208 for (q = qdisc_base; q; q = q->next) {
209 if (!strcmp(name, q->id)) {
210 if (!try_module_get(q->owner))
219 /* Set new default qdisc to use */
220 int qdisc_set_default(const char *name)
222 const struct Qdisc_ops *ops;
224 if (!capable(CAP_NET_ADMIN))
227 write_lock(&qdisc_mod_lock);
228 ops = qdisc_lookup_default(name);
230 /* Not found, drop lock and try to load module */
231 write_unlock(&qdisc_mod_lock);
232 request_module("sch_%s", name);
233 write_lock(&qdisc_mod_lock);
235 ops = qdisc_lookup_default(name);
239 /* Set new default */
240 module_put(default_qdisc_ops->owner);
241 default_qdisc_ops = ops;
243 write_unlock(&qdisc_mod_lock);
245 return ops ? 0 : -ENOENT;
248 #ifdef CONFIG_NET_SCH_DEFAULT
249 /* Set default value from kernel config */
250 static int __init sch_default_qdisc(void)
252 return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
254 late_initcall(sch_default_qdisc);
257 /* We know handle. Find qdisc among all qdisc's attached to device
258 * (root qdisc, all its children, children of children etc.)
259 * Note: caller either uses rtnl or rcu_read_lock()
262 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
266 if (!qdisc_dev(root))
267 return (root->handle == handle ? root : NULL);
269 if (!(root->flags & TCQ_F_BUILTIN) &&
270 root->handle == handle)
273 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) {
274 if (q->handle == handle)
280 void qdisc_hash_add(struct Qdisc *q, bool invisible)
282 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
284 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
286 q->flags |= TCQ_F_INVISIBLE;
289 EXPORT_SYMBOL(qdisc_hash_add);
291 void qdisc_hash_del(struct Qdisc *q)
293 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
295 hash_del_rcu(&q->hash);
298 EXPORT_SYMBOL(qdisc_hash_del);
300 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
306 q = qdisc_match_from_root(dev->qdisc, handle);
310 if (dev_ingress_queue(dev))
311 q = qdisc_match_from_root(
312 dev_ingress_queue(dev)->qdisc_sleeping,
318 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
322 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
326 cl = cops->find(p, classid);
330 leaf = cops->leaf(p, cl);
334 /* Find queueing discipline by name */
336 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
338 struct Qdisc_ops *q = NULL;
341 read_lock(&qdisc_mod_lock);
342 for (q = qdisc_base; q; q = q->next) {
343 if (nla_strcmp(kind, q->id) == 0) {
344 if (!try_module_get(q->owner))
349 read_unlock(&qdisc_mod_lock);
354 /* The linklayer setting were not transferred from iproute2, in older
355 * versions, and the rate tables lookup systems have been dropped in
356 * the kernel. To keep backward compatible with older iproute2 tc
357 * utils, we detect the linklayer setting by detecting if the rate
358 * table were modified.
360 * For linklayer ATM table entries, the rate table will be aligned to
361 * 48 bytes, thus some table entries will contain the same value. The
362 * mpu (min packet unit) is also encoded into the old rate table, thus
363 * starting from the mpu, we find low and high table entries for
364 * mapping this cell. If these entries contain the same value, when
365 * the rate tables have been modified for linklayer ATM.
367 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
368 * and then roundup to the next cell, calc the table entry one below,
371 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
373 int low = roundup(r->mpu, 48);
374 int high = roundup(low+1, 48);
375 int cell_low = low >> r->cell_log;
376 int cell_high = (high >> r->cell_log) - 1;
378 /* rtab is too inaccurate at rates > 100Mbit/s */
379 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
380 pr_debug("TC linklayer: Giving up ATM detection\n");
381 return TC_LINKLAYER_ETHERNET;
384 if ((cell_high > cell_low) && (cell_high < 256)
385 && (rtab[cell_low] == rtab[cell_high])) {
386 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
387 cell_low, cell_high, rtab[cell_high]);
388 return TC_LINKLAYER_ATM;
390 return TC_LINKLAYER_ETHERNET;
393 static struct qdisc_rate_table *qdisc_rtab_list;
395 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
397 struct netlink_ext_ack *extack)
399 struct qdisc_rate_table *rtab;
401 if (tab == NULL || r->rate == 0 ||
402 r->cell_log == 0 || r->cell_log >= 32 ||
403 nla_len(tab) != TC_RTAB_SIZE) {
404 NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
408 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
409 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
410 !memcmp(&rtab->data, nla_data(tab), 1024)) {
416 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
420 memcpy(rtab->data, nla_data(tab), 1024);
421 if (r->linklayer == TC_LINKLAYER_UNAWARE)
422 r->linklayer = __detect_linklayer(r, rtab->data);
423 rtab->next = qdisc_rtab_list;
424 qdisc_rtab_list = rtab;
426 NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
430 EXPORT_SYMBOL(qdisc_get_rtab);
432 void qdisc_put_rtab(struct qdisc_rate_table *tab)
434 struct qdisc_rate_table *rtab, **rtabp;
436 if (!tab || --tab->refcnt)
439 for (rtabp = &qdisc_rtab_list;
440 (rtab = *rtabp) != NULL;
441 rtabp = &rtab->next) {
449 EXPORT_SYMBOL(qdisc_put_rtab);
451 static LIST_HEAD(qdisc_stab_list);
453 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
454 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
455 [TCA_STAB_DATA] = { .type = NLA_BINARY },
458 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
459 struct netlink_ext_ack *extack)
461 struct nlattr *tb[TCA_STAB_MAX + 1];
462 struct qdisc_size_table *stab;
463 struct tc_sizespec *s;
464 unsigned int tsize = 0;
468 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy, extack);
471 if (!tb[TCA_STAB_BASE]) {
472 NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
473 return ERR_PTR(-EINVAL);
476 s = nla_data(tb[TCA_STAB_BASE]);
479 if (!tb[TCA_STAB_DATA]) {
480 NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
481 return ERR_PTR(-EINVAL);
483 tab = nla_data(tb[TCA_STAB_DATA]);
484 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
487 if (tsize != s->tsize || (!tab && tsize > 0)) {
488 NL_SET_ERR_MSG(extack, "Invalid size of size table");
489 return ERR_PTR(-EINVAL);
492 list_for_each_entry(stab, &qdisc_stab_list, list) {
493 if (memcmp(&stab->szopts, s, sizeof(*s)))
495 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
501 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
503 return ERR_PTR(-ENOMEM);
508 memcpy(stab->data, tab, tsize * sizeof(u16));
510 list_add_tail(&stab->list, &qdisc_stab_list);
515 static void stab_kfree_rcu(struct rcu_head *head)
517 kfree(container_of(head, struct qdisc_size_table, rcu));
520 void qdisc_put_stab(struct qdisc_size_table *tab)
525 if (--tab->refcnt == 0) {
526 list_del(&tab->list);
527 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
530 EXPORT_SYMBOL(qdisc_put_stab);
532 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
536 nest = nla_nest_start(skb, TCA_STAB);
538 goto nla_put_failure;
539 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
540 goto nla_put_failure;
541 nla_nest_end(skb, nest);
549 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
550 const struct qdisc_size_table *stab)
554 pkt_len = skb->len + stab->szopts.overhead;
555 if (unlikely(!stab->szopts.tsize))
558 slot = pkt_len + stab->szopts.cell_align;
559 if (unlikely(slot < 0))
562 slot >>= stab->szopts.cell_log;
563 if (likely(slot < stab->szopts.tsize))
564 pkt_len = stab->data[slot];
566 pkt_len = stab->data[stab->szopts.tsize - 1] *
567 (slot / stab->szopts.tsize) +
568 stab->data[slot % stab->szopts.tsize];
570 pkt_len <<= stab->szopts.size_log;
572 if (unlikely(pkt_len < 1))
574 qdisc_skb_cb(skb)->pkt_len = pkt_len;
576 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
578 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
580 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
581 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
582 txt, qdisc->ops->id, qdisc->handle >> 16);
583 qdisc->flags |= TCQ_F_WARN_NONWC;
586 EXPORT_SYMBOL(qdisc_warn_nonwc);
588 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
590 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
594 __netif_schedule(qdisc_root(wd->qdisc));
597 return HRTIMER_NORESTART;
600 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
603 hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
604 wd->timer.function = qdisc_watchdog;
607 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
609 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
611 qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
613 EXPORT_SYMBOL(qdisc_watchdog_init);
615 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
617 if (test_bit(__QDISC_STATE_DEACTIVATED,
618 &qdisc_root_sleeping(wd->qdisc)->state))
621 if (wd->last_expires == expires)
624 wd->last_expires = expires;
625 hrtimer_start(&wd->timer,
626 ns_to_ktime(expires),
627 HRTIMER_MODE_ABS_PINNED);
629 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
631 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
633 hrtimer_cancel(&wd->timer);
635 EXPORT_SYMBOL(qdisc_watchdog_cancel);
637 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
639 struct hlist_head *h;
642 h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
645 for (i = 0; i < n; i++)
646 INIT_HLIST_HEAD(&h[i]);
651 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
653 struct Qdisc_class_common *cl;
654 struct hlist_node *next;
655 struct hlist_head *nhash, *ohash;
656 unsigned int nsize, nmask, osize;
659 /* Rehash when load factor exceeds 0.75 */
660 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
662 nsize = clhash->hashsize * 2;
664 nhash = qdisc_class_hash_alloc(nsize);
668 ohash = clhash->hash;
669 osize = clhash->hashsize;
672 for (i = 0; i < osize; i++) {
673 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
674 h = qdisc_class_hash(cl->classid, nmask);
675 hlist_add_head(&cl->hnode, &nhash[h]);
678 clhash->hash = nhash;
679 clhash->hashsize = nsize;
680 clhash->hashmask = nmask;
681 sch_tree_unlock(sch);
685 EXPORT_SYMBOL(qdisc_class_hash_grow);
687 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
689 unsigned int size = 4;
691 clhash->hash = qdisc_class_hash_alloc(size);
694 clhash->hashsize = size;
695 clhash->hashmask = size - 1;
696 clhash->hashelems = 0;
699 EXPORT_SYMBOL(qdisc_class_hash_init);
701 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
703 kvfree(clhash->hash);
705 EXPORT_SYMBOL(qdisc_class_hash_destroy);
707 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
708 struct Qdisc_class_common *cl)
712 INIT_HLIST_NODE(&cl->hnode);
713 h = qdisc_class_hash(cl->classid, clhash->hashmask);
714 hlist_add_head(&cl->hnode, &clhash->hash[h]);
717 EXPORT_SYMBOL(qdisc_class_hash_insert);
719 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
720 struct Qdisc_class_common *cl)
722 hlist_del(&cl->hnode);
725 EXPORT_SYMBOL(qdisc_class_hash_remove);
727 /* Allocate an unique handle from space managed by kernel
728 * Possible range is [8000-FFFF]:0000 (0x8000 values)
730 static u32 qdisc_alloc_handle(struct net_device *dev)
733 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
736 autohandle += TC_H_MAKE(0x10000U, 0);
737 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
738 autohandle = TC_H_MAKE(0x80000000U, 0);
739 if (!qdisc_lookup(dev, autohandle))
747 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n,
750 bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
751 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 /* Notify parent qdisc only if child qdisc becomes empty.
769 * If child was empty even before update then backlog
770 * counter is screwed and we skip notification because
771 * parent class is already passive.
773 * If the original child was offloaded then it is allowed
774 * to be seem as empty, so the parent is notified anyway.
776 notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
777 !qdisc_is_offloaded);
778 /* TODO: perform the search on a per txq basis */
779 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
781 WARN_ON_ONCE(parentid != TC_H_ROOT);
784 cops = sch->ops->cl_ops;
785 if (notify && cops->qlen_notify) {
786 cl = cops->find(sch, parentid);
787 cops->qlen_notify(sch, cl);
790 sch->qstats.backlog -= len;
791 __qdisc_qstats_drop(sch, drops);
795 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
797 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
798 u32 portid, u32 seq, u16 flags, int event)
800 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
801 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
803 struct nlmsghdr *nlh;
804 unsigned char *b = skb_tail_pointer(skb);
806 struct qdisc_size_table *stab;
811 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
814 tcm = nlmsg_data(nlh);
815 tcm->tcm_family = AF_UNSPEC;
818 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
819 tcm->tcm_parent = clid;
820 tcm->tcm_handle = q->handle;
821 tcm->tcm_info = refcount_read(&q->refcnt);
822 if (nla_put_string(skb, TCA_KIND, q->ops->id))
823 goto nla_put_failure;
824 if (q->ops->ingress_block_get) {
825 block_index = q->ops->ingress_block_get(q);
827 nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
828 goto nla_put_failure;
830 if (q->ops->egress_block_get) {
831 block_index = q->ops->egress_block_get(q);
833 nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
834 goto nla_put_failure;
836 if (q->ops->dump && q->ops->dump(q, skb) < 0)
837 goto nla_put_failure;
838 if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
839 goto nla_put_failure;
840 qlen = qdisc_qlen_sum(q);
842 stab = rtnl_dereference(q->stab);
843 if (stab && qdisc_dump_stab(skb, stab) < 0)
844 goto nla_put_failure;
846 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
847 NULL, &d, TCA_PAD) < 0)
848 goto nla_put_failure;
850 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
851 goto nla_put_failure;
853 if (qdisc_is_percpu_stats(q)) {
854 cpu_bstats = q->cpu_bstats;
855 cpu_qstats = q->cpu_qstats;
858 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
859 &d, cpu_bstats, &q->bstats) < 0 ||
860 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
861 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
862 goto nla_put_failure;
864 if (gnet_stats_finish_copy(&d) < 0)
865 goto nla_put_failure;
867 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
876 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
878 if (q->flags & TCQ_F_BUILTIN)
880 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
886 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
887 struct nlmsghdr *n, u32 clid,
888 struct Qdisc *old, struct Qdisc *new)
891 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
893 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
897 if (old && !tc_qdisc_dump_ignore(old, false)) {
898 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
899 0, RTM_DELQDISC) < 0)
902 if (new && !tc_qdisc_dump_ignore(new, false)) {
903 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
904 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
909 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
910 n->nlmsg_flags & NLM_F_ECHO);
917 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
918 struct nlmsghdr *n, u32 clid,
919 struct Qdisc *old, struct Qdisc *new)
922 qdisc_notify(net, skb, n, clid, old, new);
928 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
931 * When appropriate send a netlink notification using 'skb'
934 * On success, destroy old qdisc.
937 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
938 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
939 struct Qdisc *new, struct Qdisc *old,
940 struct netlink_ext_ack *extack)
942 struct Qdisc *q = old;
943 struct net *net = dev_net(dev);
946 if (parent == NULL) {
947 unsigned int i, num_q, ingress;
950 num_q = dev->num_tx_queues;
951 if ((q && q->flags & TCQ_F_INGRESS) ||
952 (new && new->flags & TCQ_F_INGRESS)) {
955 if (!dev_ingress_queue(dev)) {
956 NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
961 if (dev->flags & IFF_UP)
964 if (new && new->ops->attach)
967 for (i = 0; i < num_q; i++) {
968 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
971 dev_queue = netdev_get_tx_queue(dev, i);
973 old = dev_graft_qdisc(dev_queue, new);
975 qdisc_refcount_inc(new);
983 notify_and_destroy(net, skb, n, classid,
985 if (new && !new->ops->attach)
986 qdisc_refcount_inc(new);
987 dev->qdisc = new ? : &noop_qdisc;
989 if (new && new->ops->attach)
990 new->ops->attach(new);
992 notify_and_destroy(net, skb, n, classid, old, new);
995 if (dev->flags & IFF_UP)
998 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1000 /* Only support running class lockless if parent is lockless */
1001 if (new && (new->flags & TCQ_F_NOLOCK) &&
1002 parent && !(parent->flags & TCQ_F_NOLOCK))
1003 new->flags &= ~TCQ_F_NOLOCK;
1006 if (cops && cops->graft) {
1007 unsigned long cl = cops->find(parent, classid);
1010 err = cops->graft(parent, cl, new, &old,
1013 NL_SET_ERR_MSG(extack, "Specified class not found");
1018 notify_and_destroy(net, skb, n, classid, old, new);
1023 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1024 struct netlink_ext_ack *extack)
1028 if (tca[TCA_INGRESS_BLOCK]) {
1029 block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1032 NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1035 if (!sch->ops->ingress_block_set) {
1036 NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1039 sch->ops->ingress_block_set(sch, block_index);
1041 if (tca[TCA_EGRESS_BLOCK]) {
1042 block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1045 NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1048 if (!sch->ops->egress_block_set) {
1049 NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1052 sch->ops->egress_block_set(sch, block_index);
1057 /* lockdep annotation is needed for ingress; egress gets it only for name */
1058 static struct lock_class_key qdisc_tx_lock;
1059 static struct lock_class_key qdisc_rx_lock;
1062 Allocate and initialize new qdisc.
1064 Parameters are passed via opt.
1067 static struct Qdisc *qdisc_create(struct net_device *dev,
1068 struct netdev_queue *dev_queue,
1069 struct Qdisc *p, u32 parent, u32 handle,
1070 struct nlattr **tca, int *errp,
1071 struct netlink_ext_ack *extack)
1074 struct nlattr *kind = tca[TCA_KIND];
1076 struct Qdisc_ops *ops;
1077 struct qdisc_size_table *stab;
1079 ops = qdisc_lookup_ops(kind);
1080 #ifdef CONFIG_MODULES
1081 if (ops == NULL && kind != NULL) {
1082 char name[IFNAMSIZ];
1083 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
1084 /* We dropped the RTNL semaphore in order to
1085 * perform the module load. So, even if we
1086 * succeeded in loading the module we have to
1087 * tell the caller to replay the request. We
1088 * indicate this using -EAGAIN.
1089 * We replay the request because the device may
1090 * go away in the mean time.
1093 request_module("sch_%s", name);
1095 ops = qdisc_lookup_ops(kind);
1097 /* We will try again qdisc_lookup_ops,
1098 * so don't keep a reference.
1100 module_put(ops->owner);
1110 NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1114 sch = qdisc_alloc(dev_queue, ops, extack);
1120 sch->parent = parent;
1122 if (handle == TC_H_INGRESS) {
1123 sch->flags |= TCQ_F_INGRESS;
1124 handle = TC_H_MAKE(TC_H_INGRESS, 0);
1125 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
1128 handle = qdisc_alloc_handle(dev);
1133 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
1134 if (!netif_is_multiqueue(dev))
1135 sch->flags |= TCQ_F_ONETXQUEUE;
1138 sch->handle = handle;
1140 /* This exist to keep backward compatible with a userspace
1141 * loophole, what allowed userspace to get IFF_NO_QUEUE
1142 * facility on older kernels by setting tx_queue_len=0 (prior
1143 * to qdisc init), and then forgot to reinit tx_queue_len
1144 * before again attaching a qdisc.
1146 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1147 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1148 netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1151 err = qdisc_block_indexes_set(sch, tca, extack);
1156 err = ops->init(sch, tca[TCA_OPTIONS], extack);
1161 if (tca[TCA_STAB]) {
1162 stab = qdisc_get_stab(tca[TCA_STAB], extack);
1164 err = PTR_ERR(stab);
1167 rcu_assign_pointer(sch->stab, stab);
1169 if (tca[TCA_RATE]) {
1170 seqcount_t *running;
1173 if (sch->flags & TCQ_F_MQROOT) {
1174 NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1178 if (sch->parent != TC_H_ROOT &&
1179 !(sch->flags & TCQ_F_INGRESS) &&
1180 (!p || !(p->flags & TCQ_F_MQROOT)))
1181 running = qdisc_root_sleeping_running(sch);
1183 running = &sch->running;
1185 err = gen_new_estimator(&sch->bstats,
1192 NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1197 qdisc_hash_add(sch, false);
1202 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1209 module_put(ops->owner);
1216 * Any broken qdiscs that would require a ops->reset() here?
1217 * The qdisc was never in action so it shouldn't be necessary.
1219 qdisc_put_stab(rtnl_dereference(sch->stab));
1225 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1226 struct netlink_ext_ack *extack)
1228 struct qdisc_size_table *ostab, *stab = NULL;
1231 if (tca[TCA_OPTIONS]) {
1232 if (!sch->ops->change) {
1233 NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1236 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1237 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1240 err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1245 if (tca[TCA_STAB]) {
1246 stab = qdisc_get_stab(tca[TCA_STAB], extack);
1248 return PTR_ERR(stab);
1251 ostab = rtnl_dereference(sch->stab);
1252 rcu_assign_pointer(sch->stab, stab);
1253 qdisc_put_stab(ostab);
1255 if (tca[TCA_RATE]) {
1256 /* NB: ignores errors from replace_estimator
1257 because change can't be undone. */
1258 if (sch->flags & TCQ_F_MQROOT)
1260 gen_replace_estimator(&sch->bstats,
1264 qdisc_root_sleeping_running(sch),
1271 struct check_loop_arg {
1272 struct qdisc_walker w;
1277 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1278 struct qdisc_walker *w);
1280 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1282 struct check_loop_arg arg;
1284 if (q->ops->cl_ops == NULL)
1287 arg.w.stop = arg.w.skip = arg.w.count = 0;
1288 arg.w.fn = check_loop_fn;
1291 q->ops->cl_ops->walk(q, &arg.w);
1292 return arg.w.stop ? -ELOOP : 0;
1296 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1299 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1300 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1302 leaf = cops->leaf(q, cl);
1304 if (leaf == arg->p || arg->depth > 7)
1306 return check_loop(leaf, arg->p, arg->depth + 1);
1311 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1312 [TCA_KIND] = { .type = NLA_NUL_STRING,
1313 .len = IFNAMSIZ - 1 },
1314 [TCA_RATE] = { .type = NLA_BINARY,
1315 .len = sizeof(struct tc_estimator) },
1316 [TCA_STAB] = { .type = NLA_NESTED },
1317 [TCA_DUMP_INVISIBLE] = { .type = NLA_FLAG },
1318 [TCA_CHAIN] = { .type = NLA_U32 },
1319 [TCA_INGRESS_BLOCK] = { .type = NLA_U32 },
1320 [TCA_EGRESS_BLOCK] = { .type = NLA_U32 },
1327 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1328 struct netlink_ext_ack *extack)
1330 struct net *net = sock_net(skb->sk);
1331 struct tcmsg *tcm = nlmsg_data(n);
1332 struct nlattr *tca[TCA_MAX + 1];
1333 struct net_device *dev;
1335 struct Qdisc *q = NULL;
1336 struct Qdisc *p = NULL;
1339 if ((n->nlmsg_type != RTM_GETQDISC) &&
1340 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1343 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, rtm_tca_policy,
1348 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1352 clid = tcm->tcm_parent;
1354 if (clid != TC_H_ROOT) {
1355 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1356 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1358 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1361 q = qdisc_leaf(p, clid);
1362 } else if (dev_ingress_queue(dev)) {
1363 q = dev_ingress_queue(dev)->qdisc_sleeping;
1369 NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1373 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1374 NL_SET_ERR_MSG(extack, "Invalid handle");
1378 q = qdisc_lookup(dev, tcm->tcm_handle);
1380 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1385 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1386 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1390 if (n->nlmsg_type == RTM_DELQDISC) {
1392 NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1395 if (q->handle == 0) {
1396 NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1399 err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1403 qdisc_notify(net, skb, n, clid, NULL, q);
1409 * Create/change qdisc.
1412 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1413 struct netlink_ext_ack *extack)
1415 struct net *net = sock_net(skb->sk);
1417 struct nlattr *tca[TCA_MAX + 1];
1418 struct net_device *dev;
1420 struct Qdisc *q, *p;
1423 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1427 /* Reinit, just in case something touches this. */
1428 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, rtm_tca_policy,
1433 tcm = nlmsg_data(n);
1434 clid = tcm->tcm_parent;
1437 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1443 if (clid != TC_H_ROOT) {
1444 if (clid != TC_H_INGRESS) {
1445 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1447 NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1450 q = qdisc_leaf(p, clid);
1451 } else if (dev_ingress_queue_create(dev)) {
1452 q = dev_ingress_queue(dev)->qdisc_sleeping;
1458 /* It may be default qdisc, ignore it */
1459 if (q && q->handle == 0)
1462 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1463 if (tcm->tcm_handle) {
1464 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1465 NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1468 if (TC_H_MIN(tcm->tcm_handle)) {
1469 NL_SET_ERR_MSG(extack, "Invalid minor handle");
1472 q = qdisc_lookup(dev, tcm->tcm_handle);
1474 goto create_n_graft;
1475 if (n->nlmsg_flags & NLM_F_EXCL) {
1476 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1479 if (tca[TCA_KIND] &&
1480 nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1481 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1485 (p && check_loop(q, p, 0))) {
1486 NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1489 qdisc_refcount_inc(q);
1493 goto create_n_graft;
1495 /* This magic test requires explanation.
1497 * We know, that some child q is already
1498 * attached to this parent and have choice:
1499 * either to change it or to create/graft new one.
1501 * 1. We are allowed to create/graft only
1502 * if CREATE and REPLACE flags are set.
1504 * 2. If EXCL is set, requestor wanted to say,
1505 * that qdisc tcm_handle is not expected
1506 * to exist, so that we choose create/graft too.
1508 * 3. The last case is when no flags are set.
1509 * Alas, it is sort of hole in API, we
1510 * cannot decide what to do unambiguously.
1511 * For now we select create/graft, if
1512 * user gave KIND, which does not match existing.
1514 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1515 (n->nlmsg_flags & NLM_F_REPLACE) &&
1516 ((n->nlmsg_flags & NLM_F_EXCL) ||
1518 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1519 goto create_n_graft;
1523 if (!tcm->tcm_handle) {
1524 NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1527 q = qdisc_lookup(dev, tcm->tcm_handle);
1530 /* Change qdisc parameters */
1532 NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1535 if (n->nlmsg_flags & NLM_F_EXCL) {
1536 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1539 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1540 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1543 err = qdisc_change(q, tca, extack);
1545 qdisc_notify(net, skb, n, clid, NULL, q);
1549 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1550 NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1553 if (clid == TC_H_INGRESS) {
1554 if (dev_ingress_queue(dev)) {
1555 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1556 tcm->tcm_parent, tcm->tcm_parent,
1559 NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1563 struct netdev_queue *dev_queue;
1565 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1566 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1568 dev_queue = p->dev_queue;
1570 dev_queue = netdev_get_tx_queue(dev, 0);
1572 q = qdisc_create(dev, dev_queue, p,
1573 tcm->tcm_parent, tcm->tcm_handle,
1583 err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1593 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1594 struct netlink_callback *cb,
1595 int *q_idx_p, int s_q_idx, bool recur,
1596 bool dump_invisible)
1598 int ret = 0, q_idx = *q_idx_p;
1606 if (q_idx < s_q_idx) {
1609 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1610 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1611 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1617 /* If dumping singletons, there is no qdisc_dev(root) and the singleton
1618 * itself has already been dumped.
1620 * If we've already dumped the top-level (ingress) qdisc above and the global
1621 * qdisc hashtable, we don't want to hit it again
1623 if (!qdisc_dev(root) || !recur)
1626 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1627 if (q_idx < s_q_idx) {
1631 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1632 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1633 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1647 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1649 struct net *net = sock_net(skb->sk);
1652 struct net_device *dev;
1653 const struct nlmsghdr *nlh = cb->nlh;
1654 struct nlattr *tca[TCA_MAX + 1];
1657 s_idx = cb->args[0];
1658 s_q_idx = q_idx = cb->args[1];
1663 err = nlmsg_parse(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1664 rtm_tca_policy, NULL);
1668 for_each_netdev(net, dev) {
1669 struct netdev_queue *dev_queue;
1677 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1678 true, tca[TCA_DUMP_INVISIBLE]) < 0)
1681 dev_queue = dev_ingress_queue(dev);
1683 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1684 &q_idx, s_q_idx, false,
1685 tca[TCA_DUMP_INVISIBLE]) < 0)
1694 cb->args[1] = q_idx;
1701 /************************************************
1702 * Traffic classes manipulation. *
1703 ************************************************/
1705 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1707 u32 portid, u32 seq, u16 flags, int event)
1710 struct nlmsghdr *nlh;
1711 unsigned char *b = skb_tail_pointer(skb);
1713 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1716 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1718 goto out_nlmsg_trim;
1719 tcm = nlmsg_data(nlh);
1720 tcm->tcm_family = AF_UNSPEC;
1723 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1724 tcm->tcm_parent = q->handle;
1725 tcm->tcm_handle = q->handle;
1727 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1728 goto nla_put_failure;
1729 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1730 goto nla_put_failure;
1732 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1733 NULL, &d, TCA_PAD) < 0)
1734 goto nla_put_failure;
1736 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1737 goto nla_put_failure;
1739 if (gnet_stats_finish_copy(&d) < 0)
1740 goto nla_put_failure;
1742 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1751 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1752 struct nlmsghdr *n, struct Qdisc *q,
1753 unsigned long cl, int event)
1755 struct sk_buff *skb;
1756 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1758 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1762 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1767 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1768 n->nlmsg_flags & NLM_F_ECHO);
1771 static int tclass_del_notify(struct net *net,
1772 const struct Qdisc_class_ops *cops,
1773 struct sk_buff *oskb, struct nlmsghdr *n,
1774 struct Qdisc *q, unsigned long cl)
1776 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1777 struct sk_buff *skb;
1783 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1787 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1788 RTM_DELTCLASS) < 0) {
1793 err = cops->delete(q, cl);
1799 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1800 n->nlmsg_flags & NLM_F_ECHO);
1803 #ifdef CONFIG_NET_CLS
1805 struct tcf_bind_args {
1806 struct tcf_walker w;
1812 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1814 struct tcf_bind_args *a = (void *)arg;
1816 if (tp->ops->bind_class) {
1817 struct Qdisc *q = tcf_block_q(tp->chain->block);
1820 tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1826 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1827 unsigned long new_cl)
1829 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1830 struct tcf_block *block;
1831 struct tcf_chain *chain;
1834 cl = cops->find(q, portid);
1837 if (!cops->tcf_block)
1839 block = cops->tcf_block(q, cl, NULL);
1842 list_for_each_entry(chain, &block->chain_list, list) {
1843 struct tcf_proto *tp;
1845 for (tp = rtnl_dereference(chain->filter_chain);
1846 tp; tp = rtnl_dereference(tp->next)) {
1847 struct tcf_bind_args arg = {};
1849 arg.w.fn = tcf_node_bind;
1853 tp->ops->walk(tp, &arg.w);
1860 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1861 unsigned long new_cl)
1867 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1868 struct netlink_ext_ack *extack)
1870 struct net *net = sock_net(skb->sk);
1871 struct tcmsg *tcm = nlmsg_data(n);
1872 struct nlattr *tca[TCA_MAX + 1];
1873 struct net_device *dev;
1874 struct Qdisc *q = NULL;
1875 const struct Qdisc_class_ops *cops;
1876 unsigned long cl = 0;
1877 unsigned long new_cl;
1883 if ((n->nlmsg_type != RTM_GETTCLASS) &&
1884 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1887 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, rtm_tca_policy,
1892 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1897 parent == TC_H_UNSPEC - unspecified parent.
1898 parent == TC_H_ROOT - class is root, which has no parent.
1899 parent == X:0 - parent is root class.
1900 parent == X:Y - parent is a node in hierarchy.
1901 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1903 handle == 0:0 - generate handle from kernel pool.
1904 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1905 handle == X:Y - clear.
1906 handle == X:0 - root class.
1909 /* Step 1. Determine qdisc handle X:0 */
1911 portid = tcm->tcm_parent;
1912 clid = tcm->tcm_handle;
1913 qid = TC_H_MAJ(clid);
1915 if (portid != TC_H_ROOT) {
1916 u32 qid1 = TC_H_MAJ(portid);
1919 /* If both majors are known, they must be identical. */
1924 } else if (qid == 0)
1925 qid = dev->qdisc->handle;
1927 /* Now qid is genuine qdisc handle consistent
1928 * both with parent and child.
1930 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1933 portid = TC_H_MAKE(qid, portid);
1936 qid = dev->qdisc->handle;
1939 /* OK. Locate qdisc */
1940 q = qdisc_lookup(dev, qid);
1944 /* An check that it supports classes */
1945 cops = q->ops->cl_ops;
1949 /* Now try to get class */
1951 if (portid == TC_H_ROOT)
1954 clid = TC_H_MAKE(qid, clid);
1957 cl = cops->find(q, clid);
1961 if (n->nlmsg_type != RTM_NEWTCLASS ||
1962 !(n->nlmsg_flags & NLM_F_CREATE))
1965 switch (n->nlmsg_type) {
1968 if (n->nlmsg_flags & NLM_F_EXCL)
1972 err = tclass_del_notify(net, cops, skb, n, q, cl);
1973 /* Unbind the class with flilters with 0 */
1974 tc_bind_tclass(q, portid, clid, 0);
1977 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1985 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1986 NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
1993 err = cops->change(q, clid, portid, tca, &new_cl, extack);
1995 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1996 /* We just create a new class, need to do reverse binding. */
1998 tc_bind_tclass(q, portid, clid, new_cl);
2004 struct qdisc_dump_args {
2005 struct qdisc_walker w;
2006 struct sk_buff *skb;
2007 struct netlink_callback *cb;
2010 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2011 struct qdisc_walker *arg)
2013 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2015 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2016 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2020 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2021 struct tcmsg *tcm, struct netlink_callback *cb,
2024 struct qdisc_dump_args arg;
2026 if (tc_qdisc_dump_ignore(q, false) ||
2027 *t_p < s_t || !q->ops->cl_ops ||
2029 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2034 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2035 arg.w.fn = qdisc_class_dump;
2039 arg.w.skip = cb->args[1];
2041 q->ops->cl_ops->walk(q, &arg.w);
2042 cb->args[1] = arg.w.count;
2049 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2050 struct tcmsg *tcm, struct netlink_callback *cb,
2051 int *t_p, int s_t, bool recur)
2059 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2062 if (!qdisc_dev(root) || !recur)
2065 if (tcm->tcm_parent) {
2066 q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2067 if (q && q != root &&
2068 tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2072 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2073 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2080 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2082 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2083 struct net *net = sock_net(skb->sk);
2084 struct netdev_queue *dev_queue;
2085 struct net_device *dev;
2088 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2090 dev = dev_get_by_index(net, tcm->tcm_ifindex);
2097 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t, true) < 0)
2100 dev_queue = dev_ingress_queue(dev);
2102 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2103 &t, s_t, false) < 0)
2113 #ifdef CONFIG_PROC_FS
2114 static int psched_show(struct seq_file *seq, void *v)
2116 seq_printf(seq, "%08x %08x %08x %08x\n",
2117 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2119 (u32)NSEC_PER_SEC / hrtimer_resolution);
2124 static int __net_init psched_net_init(struct net *net)
2126 struct proc_dir_entry *e;
2128 e = proc_create_single("psched", 0, net->proc_net, psched_show);
2135 static void __net_exit psched_net_exit(struct net *net)
2137 remove_proc_entry("psched", net->proc_net);
2140 static int __net_init psched_net_init(struct net *net)
2145 static void __net_exit psched_net_exit(struct net *net)
2150 static struct pernet_operations psched_net_ops = {
2151 .init = psched_net_init,
2152 .exit = psched_net_exit,
2155 static int __init pktsched_init(void)
2159 err = register_pernet_subsys(&psched_net_ops);
2161 pr_err("pktsched_init: "
2162 "cannot initialize per netns operations\n");
2166 register_qdisc(&pfifo_fast_ops);
2167 register_qdisc(&pfifo_qdisc_ops);
2168 register_qdisc(&bfifo_qdisc_ops);
2169 register_qdisc(&pfifo_head_drop_qdisc_ops);
2170 register_qdisc(&mq_qdisc_ops);
2171 register_qdisc(&noqueue_qdisc_ops);
2173 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2174 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2175 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2177 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2178 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2179 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2185 subsys_initcall(pktsched_init);