1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * net/sched/sch_api.c	Packet scheduler API.
   4 *
   5 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   6 *
   7 * Fixes:
   8 *
   9 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired.
  10 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  11 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support
  12 */
  13
  14#include <linux/module.h>
  15#include <linux/types.h>
  16#include <linux/kernel.h>
  17#include <linux/string.h>
  18#include <linux/errno.h>
  19#include <linux/skbuff.h>
  20#include <linux/init.h>
  21#include <linux/proc_fs.h>
  22#include <linux/seq_file.h>
  23#include <linux/kmod.h>
  24#include <linux/list.h>
  25#include <linux/hrtimer.h>
  26#include <linux/slab.h>
  27#include <linux/hashtable.h>
  28
  29#include <net/net_namespace.h>
  30#include <net/sock.h>
  31#include <net/netlink.h>
  32#include <net/pkt_sched.h>
  33#include <net/pkt_cls.h>
  34
  35#include <trace/events/qdisc.h>
  36
  37/*
  38
  39   Short review.
  40   -------------
  41
  42   This file consists of two interrelated parts:
  43
  44   1. queueing disciplines manager frontend.
  45   2. traffic classes manager frontend.
  46
  47   Generally, queueing discipline ("qdisc") is a black box,
  48   which is able to enqueue packets and to dequeue them (when
  49   device is ready to send something) in order and at times
  50   determined by algorithm hidden in it.
  51
  52   qdisc's are divided to two categories:
  53   - "queues", which have no internal structure visible from outside.
  54   - "schedulers", which split all the packets to "traffic classes",
  55     using "packet classifiers" (look at cls_api.c)
  56
  57   In turn, classes may have child qdiscs (as rule, queues)
  58   attached to them etc. etc. etc.
  59
  60   The goal of the routines in this file is to translate
  61   information supplied by user in the form of handles
  62   to more intelligible for kernel form, to make some sanity
  63   checks and part of work, which is common to all qdiscs
  64   and to provide rtnetlink notifications.
  65
  66   All real intelligent work is done inside qdisc modules.
  67
  68
  69
  70   Every discipline has two major routines: enqueue and dequeue.
  71
  72   ---dequeue
  73
  74   dequeue usually returns a skb to send. It is allowed to return NULL,
  75   but it does not mean that queue is empty, it just means that
  76   discipline does not want to send anything this time.
  77   Queue is really empty if q->q.qlen == 0.
  78   For complicated disciplines with multiple queues q->q is not
  79   real packet queue, but however q->q.qlen must be valid.
  80
  81   ---enqueue
  82
  83   enqueue returns 0, if packet was enqueued successfully.
  84   If packet (this one or another one) was dropped, it returns
  85   not zero error code.
  86   NET_XMIT_DROP 	- this packet dropped
  87     Expected action: do not backoff, but wait until queue will clear.
  88   NET_XMIT_CN	 	- probably this packet enqueued, but another one dropped.
  89     Expected action: backoff or ignore
  90
  91   Auxiliary routines:
  92
  93   ---peek
  94
  95   like dequeue but without removing a packet from the queue
  96
  97   ---reset
  98
  99   returns qdisc to initial state: purge all buffers, clear all
 100   timers, counters (except for statistics) etc.
 101
 102   ---init
 103
 104   initializes newly created qdisc.
 105
 106   ---destroy
 107
 108   destroys resources allocated by init and during lifetime of qdisc.
 109
 110   ---change
 111
 112   changes qdisc parameters.
 113 */
 114
 115/* Protects list of registered TC modules. It is pure SMP lock. */
 116static DEFINE_RWLOCK(qdisc_mod_lock);
 117
 118
 119/************************************************
 120 *	Queueing disciplines manipulation.	*
 121 ************************************************/
 122
 123
 124/* The list of all installed queueing disciplines. */
 125
 126static struct Qdisc_ops *qdisc_base;
 127
 128/* Register/unregister queueing discipline */
 129
 130int register_qdisc(struct Qdisc_ops *qops)
 131{
 132	struct Qdisc_ops *q, **qp;
 133	int rc = -EEXIST;
 134
 135	write_lock(&qdisc_mod_lock);
 136	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
 137		if (!strcmp(qops->id, q->id))
 138			goto out;
 139
 140	if (qops->enqueue == NULL)
 141		qops->enqueue = noop_qdisc_ops.enqueue;
 142	if (qops->peek == NULL) {
 143		if (qops->dequeue == NULL)
 144			qops->peek = noop_qdisc_ops.peek;
 145		else
 146			goto out_einval;
 147	}
 148	if (qops->dequeue == NULL)
 149		qops->dequeue = noop_qdisc_ops.dequeue;
 150
 151	if (qops->cl_ops) {
 152		const struct Qdisc_class_ops *cops = qops->cl_ops;
 153
 154		if (!(cops->find && cops->walk && cops->leaf))
 155			goto out_einval;
 156
 157		if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
 158			goto out_einval;
 159	}
 160
 161	qops->next = NULL;
 162	*qp = qops;
 163	rc = 0;
 164out:
 165	write_unlock(&qdisc_mod_lock);
 166	return rc;
 167
 168out_einval:
 169	rc = -EINVAL;
 170	goto out;
 171}
 172EXPORT_SYMBOL(register_qdisc);
 173
 174int unregister_qdisc(struct Qdisc_ops *qops)
 175{
 176	struct Qdisc_ops *q, **qp;
 177	int err = -ENOENT;
 178
 179	write_lock(&qdisc_mod_lock);
 180	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
 181		if (q == qops)
 182			break;
 183	if (q) {
 184		*qp = q->next;
 185		q->next = NULL;
 186		err = 0;
 187	}
 188	write_unlock(&qdisc_mod_lock);
 189	return err;
 190}
 191EXPORT_SYMBOL(unregister_qdisc);
 192
 193/* Get default qdisc if not otherwise specified */
 194void qdisc_get_default(char *name, size_t len)
 195{
 196	read_lock(&qdisc_mod_lock);
 197	strlcpy(name, default_qdisc_ops->id, len);
 198	read_unlock(&qdisc_mod_lock);
 199}
 200
 201static struct Qdisc_ops *qdisc_lookup_default(const char *name)
 202{
 203	struct Qdisc_ops *q = NULL;
 204
 205	for (q = qdisc_base; q; q = q->next) {
 206		if (!strcmp(name, q->id)) {
 207			if (!try_module_get(q->owner))
 208				q = NULL;
 209			break;
 210		}
 211	}
 212
 213	return q;
 214}
 215
 216/* Set new default qdisc to use */
 217int qdisc_set_default(const char *name)
 218{
 219	const struct Qdisc_ops *ops;
 220
 221	if (!capable(CAP_NET_ADMIN))
 222		return -EPERM;
 223
 224	write_lock(&qdisc_mod_lock);
 225	ops = qdisc_lookup_default(name);
 226	if (!ops) {
 227		/* Not found, drop lock and try to load module */
 228		write_unlock(&qdisc_mod_lock);
 229		request_module("sch_%s", name);
 230		write_lock(&qdisc_mod_lock);
 231
 232		ops = qdisc_lookup_default(name);
 233	}
 234
 235	if (ops) {
 236		/* Set new default */
 237		module_put(default_qdisc_ops->owner);
 238		default_qdisc_ops = ops;
 239	}
 240	write_unlock(&qdisc_mod_lock);
 241
 242	return ops ? 0 : -ENOENT;
 243}
 244
 245#ifdef CONFIG_NET_SCH_DEFAULT
 246/* Set default value from kernel config */
 247static int __init sch_default_qdisc(void)
 248{
 249	return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
 250}
 251late_initcall(sch_default_qdisc);
 252#endif
 253
 254/* We know handle. Find qdisc among all qdisc's attached to device
 255 * (root qdisc, all its children, children of children etc.)
 256 * Note: caller either uses rtnl or rcu_read_lock()
 257 */
 258
 259static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
 260{
 261	struct Qdisc *q;
 262
 263	if (!qdisc_dev(root))
 264		return (root->handle == handle ? root : NULL);
 265
 266	if (!(root->flags & TCQ_F_BUILTIN) &&
 267	    root->handle == handle)
 268		return root;
 269
 270	hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle,
 271				   lockdep_rtnl_is_held()) {
 272		if (q->handle == handle)
 273			return q;
 274	}
 275	return NULL;
 276}
 277
 278void qdisc_hash_add(struct Qdisc *q, bool invisible)
 279{
 280	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
 281		ASSERT_RTNL();
 282		hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
 283		if (invisible)
 284			q->flags |= TCQ_F_INVISIBLE;
 285	}
 286}
 287EXPORT_SYMBOL(qdisc_hash_add);
 288
 289void qdisc_hash_del(struct Qdisc *q)
 290{
 291	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
 292		ASSERT_RTNL();
 293		hash_del_rcu(&q->hash);
 294	}
 295}
 296EXPORT_SYMBOL(qdisc_hash_del);
 297
 298struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
 299{
 300	struct Qdisc *q;
 301
 302	if (!handle)
 303		return NULL;
 304	q = qdisc_match_from_root(dev->qdisc, handle);
 305	if (q)
 306		goto out;
 307
 308	if (dev_ingress_queue(dev))
 309		q = qdisc_match_from_root(
 310			dev_ingress_queue(dev)->qdisc_sleeping,
 311			handle);
 312out:
 313	return q;
 314}
 315
 316struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
 317{
 318	struct netdev_queue *nq;
 319	struct Qdisc *q;
 320
 321	if (!handle)
 322		return NULL;
 323	q = qdisc_match_from_root(dev->qdisc, handle);
 324	if (q)
 325		goto out;
 326
 327	nq = dev_ingress_queue_rcu(dev);
 328	if (nq)
 329		q = qdisc_match_from_root(nq->qdisc_sleeping, handle);
 330out:
 331	return q;
 332}
 333
 334static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
 335{
 336	unsigned long cl;
 337	const struct Qdisc_class_ops *cops = p->ops->cl_ops;
 338
 339	if (cops == NULL)
 340		return NULL;
 341	cl = cops->find(p, classid);
 342
 343	if (cl == 0)
 344		return NULL;
 345	return cops->leaf(p, cl);
 346}
 347
 348/* Find queueing discipline by name */
 349
 350static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
 351{
 352	struct Qdisc_ops *q = NULL;
 353
 354	if (kind) {
 355		read_lock(&qdisc_mod_lock);
 356		for (q = qdisc_base; q; q = q->next) {
 357			if (nla_strcmp(kind, q->id) == 0) {
 358				if (!try_module_get(q->owner))
 359					q = NULL;
 360				break;
 361			}
 362		}
 363		read_unlock(&qdisc_mod_lock);
 364	}
 365	return q;
 366}
 367
 368/* The linklayer setting were not transferred from iproute2, in older
 369 * versions, and the rate tables lookup systems have been dropped in
 370 * the kernel. To keep backward compatible with older iproute2 tc
 371 * utils, we detect the linklayer setting by detecting if the rate
 372 * table were modified.
 373 *
 374 * For linklayer ATM table entries, the rate table will be aligned to
 375 * 48 bytes, thus some table entries will contain the same value.  The
 376 * mpu (min packet unit) is also encoded into the old rate table, thus
 377 * starting from the mpu, we find low and high table entries for
 378 * mapping this cell.  If these entries contain the same value, when
 379 * the rate tables have been modified for linklayer ATM.
 380 *
 381 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
 382 * and then roundup to the next cell, calc the table entry one below,
 383 * and compare.
 384 */
 385static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
 386{
 387	int low       = roundup(r->mpu, 48);
 388	int high      = roundup(low+1, 48);
 389	int cell_low  = low >> r->cell_log;
 390	int cell_high = (high >> r->cell_log) - 1;
 391
 392	/* rtab is too inaccurate at rates > 100Mbit/s */
 393	if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
 394		pr_debug("TC linklayer: Giving up ATM detection\n");
 395		return TC_LINKLAYER_ETHERNET;
 396	}
 397
 398	if ((cell_high > cell_low) && (cell_high < 256)
 399	    && (rtab[cell_low] == rtab[cell_high])) {
 400		pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
 401			 cell_low, cell_high, rtab[cell_high]);
 402		return TC_LINKLAYER_ATM;
 403	}
 404	return TC_LINKLAYER_ETHERNET;
 405}
 406
 407static struct qdisc_rate_table *qdisc_rtab_list;
 408
 409struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
 410					struct nlattr *tab,
 411					struct netlink_ext_ack *extack)
 412{
 413	struct qdisc_rate_table *rtab;
 414
 415	if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
 
 416	    nla_len(tab) != TC_RTAB_SIZE) {
 417		NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
 418		return NULL;
 419	}
 420
 421	for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
 422		if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
 423		    !memcmp(&rtab->data, nla_data(tab), 1024)) {
 424			rtab->refcnt++;
 425			return rtab;
 426		}
 427	}
 428
 429	rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
 430	if (rtab) {
 431		rtab->rate = *r;
 432		rtab->refcnt = 1;
 433		memcpy(rtab->data, nla_data(tab), 1024);
 434		if (r->linklayer == TC_LINKLAYER_UNAWARE)
 435			r->linklayer = __detect_linklayer(r, rtab->data);
 436		rtab->next = qdisc_rtab_list;
 437		qdisc_rtab_list = rtab;
 438	} else {
 439		NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
 440	}
 441	return rtab;
 442}
 443EXPORT_SYMBOL(qdisc_get_rtab);
 444
 445void qdisc_put_rtab(struct qdisc_rate_table *tab)
 446{
 447	struct qdisc_rate_table *rtab, **rtabp;
 448
 449	if (!tab || --tab->refcnt)
 450		return;
 451
 452	for (rtabp = &qdisc_rtab_list;
 453	     (rtab = *rtabp) != NULL;
 454	     rtabp = &rtab->next) {
 455		if (rtab == tab) {
 456			*rtabp = rtab->next;
 457			kfree(rtab);
 458			return;
 459		}
 460	}
 461}
 462EXPORT_SYMBOL(qdisc_put_rtab);
 463
 464static LIST_HEAD(qdisc_stab_list);
 465
 466static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
 467	[TCA_STAB_BASE]	= { .len = sizeof(struct tc_sizespec) },
 468	[TCA_STAB_DATA] = { .type = NLA_BINARY },
 469};
 470
 471static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
 472					       struct netlink_ext_ack *extack)
 473{
 474	struct nlattr *tb[TCA_STAB_MAX + 1];
 475	struct qdisc_size_table *stab;
 476	struct tc_sizespec *s;
 477	unsigned int tsize = 0;
 478	u16 *tab = NULL;
 479	int err;
 480
 481	err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
 482					  extack);
 483	if (err < 0)
 484		return ERR_PTR(err);
 485	if (!tb[TCA_STAB_BASE]) {
 486		NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
 487		return ERR_PTR(-EINVAL);
 488	}
 489
 490	s = nla_data(tb[TCA_STAB_BASE]);
 491
 492	if (s->tsize > 0) {
 493		if (!tb[TCA_STAB_DATA]) {
 494			NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
 495			return ERR_PTR(-EINVAL);
 496		}
 497		tab = nla_data(tb[TCA_STAB_DATA]);
 498		tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
 499	}
 500
 501	if (tsize != s->tsize || (!tab && tsize > 0)) {
 502		NL_SET_ERR_MSG(extack, "Invalid size of size table");
 503		return ERR_PTR(-EINVAL);
 504	}
 505
 506	list_for_each_entry(stab, &qdisc_stab_list, list) {
 507		if (memcmp(&stab->szopts, s, sizeof(*s)))
 508			continue;
 509		if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
 510			continue;
 511		stab->refcnt++;
 512		return stab;
 513	}
 514
 
 
 
 
 
 
 515	stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
 516	if (!stab)
 517		return ERR_PTR(-ENOMEM);
 518
 519	stab->refcnt = 1;
 520	stab->szopts = *s;
 521	if (tsize > 0)
 522		memcpy(stab->data, tab, tsize * sizeof(u16));
 523
 524	list_add_tail(&stab->list, &qdisc_stab_list);
 525
 526	return stab;
 527}
 528
 529void qdisc_put_stab(struct qdisc_size_table *tab)
 530{
 531	if (!tab)
 532		return;
 533
 534	if (--tab->refcnt == 0) {
 535		list_del(&tab->list);
 536		kfree_rcu(tab, rcu);
 537	}
 538}
 539EXPORT_SYMBOL(qdisc_put_stab);
 540
 541static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
 542{
 543	struct nlattr *nest;
 544
 545	nest = nla_nest_start_noflag(skb, TCA_STAB);
 546	if (nest == NULL)
 547		goto nla_put_failure;
 548	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
 549		goto nla_put_failure;
 550	nla_nest_end(skb, nest);
 551
 552	return skb->len;
 553
 554nla_put_failure:
 555	return -1;
 556}
 557
 558void __qdisc_calculate_pkt_len(struct sk_buff *skb,
 559			       const struct qdisc_size_table *stab)
 560{
 561	int pkt_len, slot;
 562
 563	pkt_len = skb->len + stab->szopts.overhead;
 564	if (unlikely(!stab->szopts.tsize))
 565		goto out;
 566
 567	slot = pkt_len + stab->szopts.cell_align;
 568	if (unlikely(slot < 0))
 569		slot = 0;
 570
 571	slot >>= stab->szopts.cell_log;
 572	if (likely(slot < stab->szopts.tsize))
 573		pkt_len = stab->data[slot];
 574	else
 575		pkt_len = stab->data[stab->szopts.tsize - 1] *
 576				(slot / stab->szopts.tsize) +
 577				stab->data[slot % stab->szopts.tsize];
 578
 579	pkt_len <<= stab->szopts.size_log;
 580out:
 581	if (unlikely(pkt_len < 1))
 582		pkt_len = 1;
 583	qdisc_skb_cb(skb)->pkt_len = pkt_len;
 584}
 585EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
 586
 587void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
 588{
 589	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
 590		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
 591			txt, qdisc->ops->id, qdisc->handle >> 16);
 592		qdisc->flags |= TCQ_F_WARN_NONWC;
 593	}
 594}
 595EXPORT_SYMBOL(qdisc_warn_nonwc);
 596
 597static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
 598{
 599	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
 600						 timer);
 601
 602	rcu_read_lock();
 603	__netif_schedule(qdisc_root(wd->qdisc));
 604	rcu_read_unlock();
 605
 606	return HRTIMER_NORESTART;
 607}
 608
 609void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
 610				 clockid_t clockid)
 611{
 612	hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
 613	wd->timer.function = qdisc_watchdog;
 614	wd->qdisc = qdisc;
 615}
 616EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
 617
 618void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
 619{
 620	qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
 621}
 622EXPORT_SYMBOL(qdisc_watchdog_init);
 623
 624void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
 625				      u64 delta_ns)
 626{
 627	if (test_bit(__QDISC_STATE_DEACTIVATED,
 628		     &qdisc_root_sleeping(wd->qdisc)->state))
 629		return;
 630
 631	if (hrtimer_is_queued(&wd->timer)) {
 632		/* If timer is already set in [expires, expires + delta_ns],
 633		 * do not reprogram it.
 634		 */
 635		if (wd->last_expires - expires <= delta_ns)
 636			return;
 637	}
 638
 639	wd->last_expires = expires;
 640	hrtimer_start_range_ns(&wd->timer,
 641			       ns_to_ktime(expires),
 642			       delta_ns,
 643			       HRTIMER_MODE_ABS_PINNED);
 644}
 645EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
 646
 647void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
 648{
 649	hrtimer_cancel(&wd->timer);
 650}
 651EXPORT_SYMBOL(qdisc_watchdog_cancel);
 652
 653static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
 654{
 655	struct hlist_head *h;
 656	unsigned int i;
 657
 658	h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
 659
 660	if (h != NULL) {
 661		for (i = 0; i < n; i++)
 662			INIT_HLIST_HEAD(&h[i]);
 663	}
 664	return h;
 665}
 666
 667void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
 668{
 669	struct Qdisc_class_common *cl;
 670	struct hlist_node *next;
 671	struct hlist_head *nhash, *ohash;
 672	unsigned int nsize, nmask, osize;
 673	unsigned int i, h;
 674
 675	/* Rehash when load factor exceeds 0.75 */
 676	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
 677		return;
 678	nsize = clhash->hashsize * 2;
 679	nmask = nsize - 1;
 680	nhash = qdisc_class_hash_alloc(nsize);
 681	if (nhash == NULL)
 682		return;
 683
 684	ohash = clhash->hash;
 685	osize = clhash->hashsize;
 686
 687	sch_tree_lock(sch);
 688	for (i = 0; i < osize; i++) {
 689		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
 690			h = qdisc_class_hash(cl->classid, nmask);
 691			hlist_add_head(&cl->hnode, &nhash[h]);
 692		}
 693	}
 694	clhash->hash     = nhash;
 695	clhash->hashsize = nsize;
 696	clhash->hashmask = nmask;
 697	sch_tree_unlock(sch);
 698
 699	kvfree(ohash);
 700}
 701EXPORT_SYMBOL(qdisc_class_hash_grow);
 702
 703int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
 704{
 705	unsigned int size = 4;
 706
 707	clhash->hash = qdisc_class_hash_alloc(size);
 708	if (!clhash->hash)
 709		return -ENOMEM;
 710	clhash->hashsize  = size;
 711	clhash->hashmask  = size - 1;
 712	clhash->hashelems = 0;
 713	return 0;
 714}
 715EXPORT_SYMBOL(qdisc_class_hash_init);
 716
 717void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
 718{
 719	kvfree(clhash->hash);
 720}
 721EXPORT_SYMBOL(qdisc_class_hash_destroy);
 722
 723void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
 724			     struct Qdisc_class_common *cl)
 725{
 726	unsigned int h;
 727
 728	INIT_HLIST_NODE(&cl->hnode);
 729	h = qdisc_class_hash(cl->classid, clhash->hashmask);
 730	hlist_add_head(&cl->hnode, &clhash->hash[h]);
 731	clhash->hashelems++;
 732}
 733EXPORT_SYMBOL(qdisc_class_hash_insert);
 734
 735void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
 736			     struct Qdisc_class_common *cl)
 737{
 738	hlist_del(&cl->hnode);
 739	clhash->hashelems--;
 740}
 741EXPORT_SYMBOL(qdisc_class_hash_remove);
 742
 743/* Allocate an unique handle from space managed by kernel
 744 * Possible range is [8000-FFFF]:0000 (0x8000 values)
 745 */
 746static u32 qdisc_alloc_handle(struct net_device *dev)
 747{
 748	int i = 0x8000;
 749	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
 750
 751	do {
 752		autohandle += TC_H_MAKE(0x10000U, 0);
 753		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
 754			autohandle = TC_H_MAKE(0x80000000U, 0);
 755		if (!qdisc_lookup(dev, autohandle))
 756			return autohandle;
 757		cond_resched();
 758	} while	(--i > 0);
 759
 760	return 0;
 761}
 762
 763void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
 764{
 765	bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
 766	const struct Qdisc_class_ops *cops;
 767	unsigned long cl;
 768	u32 parentid;
 769	bool notify;
 770	int drops;
 771
 772	if (n == 0 && len == 0)
 773		return;
 774	drops = max_t(int, n, 0);
 775	rcu_read_lock();
 776	while ((parentid = sch->parent)) {
 777		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
 778			break;
 779
 780		if (sch->flags & TCQ_F_NOPARENT)
 781			break;
 782		/* Notify parent qdisc only if child qdisc becomes empty.
 783		 *
 784		 * If child was empty even before update then backlog
 785		 * counter is screwed and we skip notification because
 786		 * parent class is already passive.
 787		 *
 788		 * If the original child was offloaded then it is allowed
 789		 * to be seem as empty, so the parent is notified anyway.
 790		 */
 791		notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
 792						       !qdisc_is_offloaded);
 793		/* TODO: perform the search on a per txq basis */
 794		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
 795		if (sch == NULL) {
 796			WARN_ON_ONCE(parentid != TC_H_ROOT);
 797			break;
 798		}
 799		cops = sch->ops->cl_ops;
 800		if (notify && cops->qlen_notify) {
 801			cl = cops->find(sch, parentid);
 802			cops->qlen_notify(sch, cl);
 803		}
 804		sch->q.qlen -= n;
 805		sch->qstats.backlog -= len;
 806		__qdisc_qstats_drop(sch, drops);
 807	}
 808	rcu_read_unlock();
 809}
 810EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
 811
 812int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
 813			      void *type_data)
 814{
 815	struct net_device *dev = qdisc_dev(sch);
 816	int err;
 817
 818	sch->flags &= ~TCQ_F_OFFLOADED;
 819	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
 820		return 0;
 821
 822	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
 823	if (err == -EOPNOTSUPP)
 824		return 0;
 825
 826	if (!err)
 827		sch->flags |= TCQ_F_OFFLOADED;
 828
 829	return err;
 830}
 831EXPORT_SYMBOL(qdisc_offload_dump_helper);
 832
 833void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
 834				struct Qdisc *new, struct Qdisc *old,
 835				enum tc_setup_type type, void *type_data,
 836				struct netlink_ext_ack *extack)
 837{
 838	bool any_qdisc_is_offloaded;
 839	int err;
 840
 841	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
 842		return;
 843
 844	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
 845
 846	/* Don't report error if the graft is part of destroy operation. */
 847	if (!err || !new || new == &noop_qdisc)
 848		return;
 849
 850	/* Don't report error if the parent, the old child and the new
 851	 * one are not offloaded.
 852	 */
 853	any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
 854	any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
 855	any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
 856
 857	if (any_qdisc_is_offloaded)
 858		NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
 859}
 860EXPORT_SYMBOL(qdisc_offload_graft_helper);
 861
 862static void qdisc_offload_graft_root(struct net_device *dev,
 863				     struct Qdisc *new, struct Qdisc *old,
 864				     struct netlink_ext_ack *extack)
 865{
 866	struct tc_root_qopt_offload graft_offload = {
 867		.command	= TC_ROOT_GRAFT,
 868		.handle		= new ? new->handle : 0,
 869		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
 870				  (old && old->flags & TCQ_F_INGRESS),
 871	};
 872
 873	qdisc_offload_graft_helper(dev, NULL, new, old,
 874				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
 875}
 876
 877static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
 878			 u32 portid, u32 seq, u16 flags, int event)
 879{
 880	struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
 881	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
 882	struct tcmsg *tcm;
 883	struct nlmsghdr  *nlh;
 884	unsigned char *b = skb_tail_pointer(skb);
 885	struct gnet_dump d;
 886	struct qdisc_size_table *stab;
 887	u32 block_index;
 888	__u32 qlen;
 889
 890	cond_resched();
 891	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
 892	if (!nlh)
 893		goto out_nlmsg_trim;
 894	tcm = nlmsg_data(nlh);
 895	tcm->tcm_family = AF_UNSPEC;
 896	tcm->tcm__pad1 = 0;
 897	tcm->tcm__pad2 = 0;
 898	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
 899	tcm->tcm_parent = clid;
 900	tcm->tcm_handle = q->handle;
 901	tcm->tcm_info = refcount_read(&q->refcnt);
 902	if (nla_put_string(skb, TCA_KIND, q->ops->id))
 903		goto nla_put_failure;
 904	if (q->ops->ingress_block_get) {
 905		block_index = q->ops->ingress_block_get(q);
 906		if (block_index &&
 907		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
 908			goto nla_put_failure;
 909	}
 910	if (q->ops->egress_block_get) {
 911		block_index = q->ops->egress_block_get(q);
 912		if (block_index &&
 913		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
 914			goto nla_put_failure;
 915	}
 916	if (q->ops->dump && q->ops->dump(q, skb) < 0)
 917		goto nla_put_failure;
 918	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
 919		goto nla_put_failure;
 920	qlen = qdisc_qlen_sum(q);
 921
 922	stab = rtnl_dereference(q->stab);
 923	if (stab && qdisc_dump_stab(skb, stab) < 0)
 924		goto nla_put_failure;
 925
 926	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
 927					 NULL, &d, TCA_PAD) < 0)
 928		goto nla_put_failure;
 929
 930	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
 931		goto nla_put_failure;
 932
 933	if (qdisc_is_percpu_stats(q)) {
 934		cpu_bstats = q->cpu_bstats;
 935		cpu_qstats = q->cpu_qstats;
 936	}
 937
 938	if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
 939				  &d, cpu_bstats, &q->bstats) < 0 ||
 940	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
 941	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
 942		goto nla_put_failure;
 943
 944	if (gnet_stats_finish_copy(&d) < 0)
 945		goto nla_put_failure;
 946
 947	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 948	return skb->len;
 949
 950out_nlmsg_trim:
 951nla_put_failure:
 952	nlmsg_trim(skb, b);
 953	return -1;
 954}
 955
 956static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
 957{
 958	if (q->flags & TCQ_F_BUILTIN)
 959		return true;
 960	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
 961		return true;
 962
 963	return false;
 964}
 965
 966static int qdisc_notify(struct net *net, struct sk_buff *oskb,
 967			struct nlmsghdr *n, u32 clid,
 968			struct Qdisc *old, struct Qdisc *new)
 969{
 970	struct sk_buff *skb;
 971	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
 972
 973	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 974	if (!skb)
 975		return -ENOBUFS;
 976
 977	if (old && !tc_qdisc_dump_ignore(old, false)) {
 978		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
 979				  0, RTM_DELQDISC) < 0)
 980			goto err_out;
 981	}
 982	if (new && !tc_qdisc_dump_ignore(new, false)) {
 983		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
 984				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
 985			goto err_out;
 986	}
 987
 988	if (skb->len)
 989		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 990				      n->nlmsg_flags & NLM_F_ECHO);
 991
 992err_out:
 993	kfree_skb(skb);
 994	return -EINVAL;
 995}
 996
 997static void notify_and_destroy(struct net *net, struct sk_buff *skb,
 998			       struct nlmsghdr *n, u32 clid,
 999			       struct Qdisc *old, struct Qdisc *new)
1000{
1001	if (new || old)
1002		qdisc_notify(net, skb, n, clid, old, new);
1003
1004	if (old)
1005		qdisc_put(old);
1006}
1007
1008static void qdisc_clear_nolock(struct Qdisc *sch)
1009{
1010	sch->flags &= ~TCQ_F_NOLOCK;
1011	if (!(sch->flags & TCQ_F_CPUSTATS))
1012		return;
1013
1014	free_percpu(sch->cpu_bstats);
1015	free_percpu(sch->cpu_qstats);
1016	sch->cpu_bstats = NULL;
1017	sch->cpu_qstats = NULL;
1018	sch->flags &= ~TCQ_F_CPUSTATS;
1019}
1020
1021/* Graft qdisc "new" to class "classid" of qdisc "parent" or
1022 * to device "dev".
1023 *
1024 * When appropriate send a netlink notification using 'skb'
1025 * and "n".
1026 *
1027 * On success, destroy old qdisc.
1028 */
1029
1030static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1031		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1032		       struct Qdisc *new, struct Qdisc *old,
1033		       struct netlink_ext_ack *extack)
1034{
1035	struct Qdisc *q = old;
1036	struct net *net = dev_net(dev);
1037
1038	if (parent == NULL) {
1039		unsigned int i, num_q, ingress;
1040
1041		ingress = 0;
1042		num_q = dev->num_tx_queues;
1043		if ((q && q->flags & TCQ_F_INGRESS) ||
1044		    (new && new->flags & TCQ_F_INGRESS)) {
1045			num_q = 1;
1046			ingress = 1;
1047			if (!dev_ingress_queue(dev)) {
1048				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1049				return -ENOENT;
1050			}
1051		}
1052
1053		if (dev->flags & IFF_UP)
1054			dev_deactivate(dev);
1055
1056		qdisc_offload_graft_root(dev, new, old, extack);
1057
1058		if (new && new->ops->attach)
1059			goto skip;
1060
1061		for (i = 0; i < num_q; i++) {
1062			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1063
1064			if (!ingress)
1065				dev_queue = netdev_get_tx_queue(dev, i);
1066
1067			old = dev_graft_qdisc(dev_queue, new);
1068			if (new && i > 0)
1069				qdisc_refcount_inc(new);
1070
1071			if (!ingress)
1072				qdisc_put(old);
1073		}
1074
1075skip:
1076		if (!ingress) {
1077			notify_and_destroy(net, skb, n, classid,
1078					   dev->qdisc, new);
1079			if (new && !new->ops->attach)
1080				qdisc_refcount_inc(new);
1081			dev->qdisc = new ? : &noop_qdisc;
1082
1083			if (new && new->ops->attach)
1084				new->ops->attach(new);
1085		} else {
1086			notify_and_destroy(net, skb, n, classid, old, new);
1087		}
1088
1089		if (dev->flags & IFF_UP)
1090			dev_activate(dev);
1091	} else {
1092		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1093		unsigned long cl;
1094		int err;
1095
1096		/* Only support running class lockless if parent is lockless */
1097		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1098			qdisc_clear_nolock(new);
1099
1100		if (!cops || !cops->graft)
1101			return -EOPNOTSUPP;
1102
1103		cl = cops->find(parent, classid);
1104		if (!cl) {
1105			NL_SET_ERR_MSG(extack, "Specified class not found");
1106			return -ENOENT;
1107		}
1108
1109		err = cops->graft(parent, cl, new, &old, extack);
1110		if (err)
1111			return err;
1112		notify_and_destroy(net, skb, n, classid, old, new);
1113	}
1114	return 0;
1115}
1116
1117static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1118				   struct netlink_ext_ack *extack)
1119{
1120	u32 block_index;
1121
1122	if (tca[TCA_INGRESS_BLOCK]) {
1123		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1124
1125		if (!block_index) {
1126			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1127			return -EINVAL;
1128		}
1129		if (!sch->ops->ingress_block_set) {
1130			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1131			return -EOPNOTSUPP;
1132		}
1133		sch->ops->ingress_block_set(sch, block_index);
1134	}
1135	if (tca[TCA_EGRESS_BLOCK]) {
1136		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1137
1138		if (!block_index) {
1139			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1140			return -EINVAL;
1141		}
1142		if (!sch->ops->egress_block_set) {
1143			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1144			return -EOPNOTSUPP;
1145		}
1146		sch->ops->egress_block_set(sch, block_index);
1147	}
1148	return 0;
1149}
1150
1151/*
1152   Allocate and initialize new qdisc.
1153
1154   Parameters are passed via opt.
1155 */
1156
1157static struct Qdisc *qdisc_create(struct net_device *dev,
1158				  struct netdev_queue *dev_queue,
1159				  struct Qdisc *p, u32 parent, u32 handle,
1160				  struct nlattr **tca, int *errp,
1161				  struct netlink_ext_ack *extack)
1162{
1163	int err;
1164	struct nlattr *kind = tca[TCA_KIND];
1165	struct Qdisc *sch;
1166	struct Qdisc_ops *ops;
1167	struct qdisc_size_table *stab;
1168
1169	ops = qdisc_lookup_ops(kind);
1170#ifdef CONFIG_MODULES
1171	if (ops == NULL && kind != NULL) {
1172		char name[IFNAMSIZ];
1173		if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
1174			/* We dropped the RTNL semaphore in order to
1175			 * perform the module load.  So, even if we
1176			 * succeeded in loading the module we have to
1177			 * tell the caller to replay the request.  We
1178			 * indicate this using -EAGAIN.
1179			 * We replay the request because the device may
1180			 * go away in the mean time.
1181			 */
1182			rtnl_unlock();
1183			request_module("sch_%s", name);
1184			rtnl_lock();
1185			ops = qdisc_lookup_ops(kind);
1186			if (ops != NULL) {
1187				/* We will try again qdisc_lookup_ops,
1188				 * so don't keep a reference.
1189				 */
1190				module_put(ops->owner);
1191				err = -EAGAIN;
1192				goto err_out;
1193			}
1194		}
1195	}
1196#endif
1197
1198	err = -ENOENT;
1199	if (!ops) {
1200		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1201		goto err_out;
1202	}
1203
1204	sch = qdisc_alloc(dev_queue, ops, extack);
1205	if (IS_ERR(sch)) {
1206		err = PTR_ERR(sch);
1207		goto err_out2;
1208	}
1209
1210	sch->parent = parent;
1211
1212	if (handle == TC_H_INGRESS) {
1213		sch->flags |= TCQ_F_INGRESS;
1214		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1215	} else {
1216		if (handle == 0) {
1217			handle = qdisc_alloc_handle(dev);
1218			if (handle == 0) {
1219				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1220				err = -ENOSPC;
1221				goto err_out3;
1222			}
1223		}
1224		if (!netif_is_multiqueue(dev))
1225			sch->flags |= TCQ_F_ONETXQUEUE;
1226	}
1227
1228	sch->handle = handle;
1229
1230	/* This exist to keep backward compatible with a userspace
1231	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1232	 * facility on older kernels by setting tx_queue_len=0 (prior
1233	 * to qdisc init), and then forgot to reinit tx_queue_len
1234	 * before again attaching a qdisc.
1235	 */
1236	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1237		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1238		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1239	}
1240
1241	err = qdisc_block_indexes_set(sch, tca, extack);
1242	if (err)
1243		goto err_out3;
1244
1245	if (ops->init) {
1246		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1247		if (err != 0)
1248			goto err_out5;
1249	}
1250
1251	if (tca[TCA_STAB]) {
1252		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1253		if (IS_ERR(stab)) {
1254			err = PTR_ERR(stab);
1255			goto err_out4;
1256		}
1257		rcu_assign_pointer(sch->stab, stab);
1258	}
1259	if (tca[TCA_RATE]) {
1260		seqcount_t *running;
1261
1262		err = -EOPNOTSUPP;
1263		if (sch->flags & TCQ_F_MQROOT) {
1264			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1265			goto err_out4;
1266		}
1267
1268		if (sch->parent != TC_H_ROOT &&
1269		    !(sch->flags & TCQ_F_INGRESS) &&
1270		    (!p || !(p->flags & TCQ_F_MQROOT)))
1271			running = qdisc_root_sleeping_running(sch);
1272		else
1273			running = &sch->running;
1274
1275		err = gen_new_estimator(&sch->bstats,
1276					sch->cpu_bstats,
1277					&sch->rate_est,
1278					NULL,
1279					running,
1280					tca[TCA_RATE]);
1281		if (err) {
1282			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1283			goto err_out4;
1284		}
1285	}
1286
1287	qdisc_hash_add(sch, false);
1288	trace_qdisc_create(ops, dev, parent);
1289
1290	return sch;
1291
1292err_out5:
1293	/* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1294	if (ops->destroy)
1295		ops->destroy(sch);
1296err_out3:
1297	dev_put(dev);
1298	qdisc_free(sch);
1299err_out2:
1300	module_put(ops->owner);
1301err_out:
1302	*errp = err;
1303	return NULL;
1304
1305err_out4:
1306	/*
1307	 * Any broken qdiscs that would require a ops->reset() here?
1308	 * The qdisc was never in action so it shouldn't be necessary.
1309	 */
1310	qdisc_put_stab(rtnl_dereference(sch->stab));
1311	if (ops->destroy)
1312		ops->destroy(sch);
1313	goto err_out3;
1314}
1315
1316static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1317			struct netlink_ext_ack *extack)
1318{
1319	struct qdisc_size_table *ostab, *stab = NULL;
1320	int err = 0;
1321
1322	if (tca[TCA_OPTIONS]) {
1323		if (!sch->ops->change) {
1324			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1325			return -EINVAL;
1326		}
1327		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1328			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1329			return -EOPNOTSUPP;
1330		}
1331		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1332		if (err)
1333			return err;
1334	}
1335
1336	if (tca[TCA_STAB]) {
1337		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1338		if (IS_ERR(stab))
1339			return PTR_ERR(stab);
1340	}
1341
1342	ostab = rtnl_dereference(sch->stab);
1343	rcu_assign_pointer(sch->stab, stab);
1344	qdisc_put_stab(ostab);
1345
1346	if (tca[TCA_RATE]) {
1347		/* NB: ignores errors from replace_estimator
1348		   because change can't be undone. */
1349		if (sch->flags & TCQ_F_MQROOT)
1350			goto out;
1351		gen_replace_estimator(&sch->bstats,
1352				      sch->cpu_bstats,
1353				      &sch->rate_est,
1354				      NULL,
1355				      qdisc_root_sleeping_running(sch),
1356				      tca[TCA_RATE]);
1357	}
1358out:
1359	return 0;
1360}
1361
1362struct check_loop_arg {
1363	struct qdisc_walker	w;
1364	struct Qdisc		*p;
1365	int			depth;
1366};
1367
1368static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1369			 struct qdisc_walker *w);
1370
1371static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1372{
1373	struct check_loop_arg	arg;
1374
1375	if (q->ops->cl_ops == NULL)
1376		return 0;
1377
1378	arg.w.stop = arg.w.skip = arg.w.count = 0;
1379	arg.w.fn = check_loop_fn;
1380	arg.depth = depth;
1381	arg.p = p;
1382	q->ops->cl_ops->walk(q, &arg.w);
1383	return arg.w.stop ? -ELOOP : 0;
1384}
1385
1386static int
1387check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1388{
1389	struct Qdisc *leaf;
1390	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1391	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1392
1393	leaf = cops->leaf(q, cl);
1394	if (leaf) {
1395		if (leaf == arg->p || arg->depth > 7)
1396			return -ELOOP;
1397		return check_loop(leaf, arg->p, arg->depth + 1);
1398	}
1399	return 0;
1400}
1401
1402const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1403	[TCA_KIND]		= { .type = NLA_STRING },
1404	[TCA_RATE]		= { .type = NLA_BINARY,
1405				    .len = sizeof(struct tc_estimator) },
1406	[TCA_STAB]		= { .type = NLA_NESTED },
1407	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1408	[TCA_CHAIN]		= { .type = NLA_U32 },
1409	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1410	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1411};
1412
1413/*
1414 * Delete/get qdisc.
1415 */
1416
1417static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1418			struct netlink_ext_ack *extack)
1419{
1420	struct net *net = sock_net(skb->sk);
1421	struct tcmsg *tcm = nlmsg_data(n);
1422	struct nlattr *tca[TCA_MAX + 1];
1423	struct net_device *dev;
1424	u32 clid;
1425	struct Qdisc *q = NULL;
1426	struct Qdisc *p = NULL;
1427	int err;
1428
1429	if ((n->nlmsg_type != RTM_GETQDISC) &&
1430	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1431		return -EPERM;
1432
1433	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1434				     rtm_tca_policy, extack);
1435	if (err < 0)
1436		return err;
1437
1438	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1439	if (!dev)
1440		return -ENODEV;
1441
1442	clid = tcm->tcm_parent;
1443	if (clid) {
1444		if (clid != TC_H_ROOT) {
1445			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1446				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1447				if (!p) {
1448					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1449					return -ENOENT;
1450				}
1451				q = qdisc_leaf(p, clid);
1452			} else if (dev_ingress_queue(dev)) {
1453				q = dev_ingress_queue(dev)->qdisc_sleeping;
1454			}
1455		} else {
1456			q = dev->qdisc;
1457		}
1458		if (!q) {
1459			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1460			return -ENOENT;
1461		}
1462
1463		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1464			NL_SET_ERR_MSG(extack, "Invalid handle");
1465			return -EINVAL;
1466		}
1467	} else {
1468		q = qdisc_lookup(dev, tcm->tcm_handle);
1469		if (!q) {
1470			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1471			return -ENOENT;
1472		}
1473	}
1474
1475	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1476		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1477		return -EINVAL;
1478	}
1479
1480	if (n->nlmsg_type == RTM_DELQDISC) {
1481		if (!clid) {
1482			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1483			return -EINVAL;
1484		}
1485		if (q->handle == 0) {
1486			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1487			return -ENOENT;
1488		}
1489		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1490		if (err != 0)
1491			return err;
1492	} else {
1493		qdisc_notify(net, skb, n, clid, NULL, q);
1494	}
1495	return 0;
1496}
1497
1498/*
1499 * Create/change qdisc.
1500 */
1501
1502static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1503			   struct netlink_ext_ack *extack)
1504{
1505	struct net *net = sock_net(skb->sk);
1506	struct tcmsg *tcm;
1507	struct nlattr *tca[TCA_MAX + 1];
1508	struct net_device *dev;
1509	u32 clid;
1510	struct Qdisc *q, *p;
1511	int err;
1512
1513	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1514		return -EPERM;
1515
1516replay:
1517	/* Reinit, just in case something touches this. */
1518	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1519				     rtm_tca_policy, extack);
1520	if (err < 0)
1521		return err;
1522
1523	tcm = nlmsg_data(n);
1524	clid = tcm->tcm_parent;
1525	q = p = NULL;
1526
1527	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1528	if (!dev)
1529		return -ENODEV;
1530
1531
1532	if (clid) {
1533		if (clid != TC_H_ROOT) {
1534			if (clid != TC_H_INGRESS) {
1535				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1536				if (!p) {
1537					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1538					return -ENOENT;
1539				}
1540				q = qdisc_leaf(p, clid);
1541			} else if (dev_ingress_queue_create(dev)) {
1542				q = dev_ingress_queue(dev)->qdisc_sleeping;
1543			}
1544		} else {
1545			q = dev->qdisc;
1546		}
1547
1548		/* It may be default qdisc, ignore it */
1549		if (q && q->handle == 0)
1550			q = NULL;
1551
1552		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1553			if (tcm->tcm_handle) {
1554				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1555					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1556					return -EEXIST;
1557				}
1558				if (TC_H_MIN(tcm->tcm_handle)) {
1559					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1560					return -EINVAL;
1561				}
1562				q = qdisc_lookup(dev, tcm->tcm_handle);
1563				if (!q)
1564					goto create_n_graft;
1565				if (n->nlmsg_flags & NLM_F_EXCL) {
1566					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1567					return -EEXIST;
1568				}
1569				if (tca[TCA_KIND] &&
1570				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1571					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1572					return -EINVAL;
1573				}
1574				if (q == p ||
1575				    (p && check_loop(q, p, 0))) {
1576					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1577					return -ELOOP;
1578				}
1579				qdisc_refcount_inc(q);
1580				goto graft;
1581			} else {
1582				if (!q)
1583					goto create_n_graft;
1584
1585				/* This magic test requires explanation.
1586				 *
1587				 *   We know, that some child q is already
1588				 *   attached to this parent and have choice:
1589				 *   either to change it or to create/graft new one.
1590				 *
1591				 *   1. We are allowed to create/graft only
1592				 *   if CREATE and REPLACE flags are set.
1593				 *
1594				 *   2. If EXCL is set, requestor wanted to say,
1595				 *   that qdisc tcm_handle is not expected
1596				 *   to exist, so that we choose create/graft too.
1597				 *
1598				 *   3. The last case is when no flags are set.
1599				 *   Alas, it is sort of hole in API, we
1600				 *   cannot decide what to do unambiguously.
1601				 *   For now we select create/graft, if
1602				 *   user gave KIND, which does not match existing.
1603				 */
1604				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1605				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1606				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1607				     (tca[TCA_KIND] &&
1608				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1609					goto create_n_graft;
1610			}
1611		}
1612	} else {
1613		if (!tcm->tcm_handle) {
1614			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1615			return -EINVAL;
1616		}
1617		q = qdisc_lookup(dev, tcm->tcm_handle);
1618	}
1619
1620	/* Change qdisc parameters */
1621	if (!q) {
1622		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1623		return -ENOENT;
1624	}
1625	if (n->nlmsg_flags & NLM_F_EXCL) {
1626		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1627		return -EEXIST;
1628	}
1629	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1630		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1631		return -EINVAL;
1632	}
1633	err = qdisc_change(q, tca, extack);
1634	if (err == 0)
1635		qdisc_notify(net, skb, n, clid, NULL, q);
1636	return err;
1637
1638create_n_graft:
1639	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1640		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1641		return -ENOENT;
1642	}
1643	if (clid == TC_H_INGRESS) {
1644		if (dev_ingress_queue(dev)) {
1645			q = qdisc_create(dev, dev_ingress_queue(dev), p,
1646					 tcm->tcm_parent, tcm->tcm_parent,
1647					 tca, &err, extack);
1648		} else {
1649			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1650			err = -ENOENT;
1651		}
1652	} else {
1653		struct netdev_queue *dev_queue;
1654
1655		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1656			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1657		else if (p)
1658			dev_queue = p->dev_queue;
1659		else
1660			dev_queue = netdev_get_tx_queue(dev, 0);
1661
1662		q = qdisc_create(dev, dev_queue, p,
1663				 tcm->tcm_parent, tcm->tcm_handle,
1664				 tca, &err, extack);
1665	}
1666	if (q == NULL) {
1667		if (err == -EAGAIN)
1668			goto replay;
1669		return err;
1670	}
1671
1672graft:
1673	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1674	if (err) {
1675		if (q)
1676			qdisc_put(q);
1677		return err;
1678	}
1679
1680	return 0;
1681}
1682
1683static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1684			      struct netlink_callback *cb,
1685			      int *q_idx_p, int s_q_idx, bool recur,
1686			      bool dump_invisible)
1687{
1688	int ret = 0, q_idx = *q_idx_p;
1689	struct Qdisc *q;
1690	int b;
1691
1692	if (!root)
1693		return 0;
1694
1695	q = root;
1696	if (q_idx < s_q_idx) {
1697		q_idx++;
1698	} else {
1699		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1700		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1701				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1702				  RTM_NEWQDISC) <= 0)
1703			goto done;
1704		q_idx++;
1705	}
1706
1707	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1708	 * itself has already been dumped.
1709	 *
1710	 * If we've already dumped the top-level (ingress) qdisc above and the global
1711	 * qdisc hashtable, we don't want to hit it again
1712	 */
1713	if (!qdisc_dev(root) || !recur)
1714		goto out;
1715
1716	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1717		if (q_idx < s_q_idx) {
1718			q_idx++;
1719			continue;
1720		}
1721		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1722		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1723				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1724				  RTM_NEWQDISC) <= 0)
1725			goto done;
1726		q_idx++;
1727	}
1728
1729out:
1730	*q_idx_p = q_idx;
1731	return ret;
1732done:
1733	ret = -1;
1734	goto out;
1735}
1736
1737static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1738{
1739	struct net *net = sock_net(skb->sk);
1740	int idx, q_idx;
1741	int s_idx, s_q_idx;
1742	struct net_device *dev;
1743	const struct nlmsghdr *nlh = cb->nlh;
1744	struct nlattr *tca[TCA_MAX + 1];
1745	int err;
1746
1747	s_idx = cb->args[0];
1748	s_q_idx = q_idx = cb->args[1];
1749
1750	idx = 0;
1751	ASSERT_RTNL();
1752
1753	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1754				     rtm_tca_policy, cb->extack);
1755	if (err < 0)
1756		return err;
1757
1758	for_each_netdev(net, dev) {
1759		struct netdev_queue *dev_queue;
1760
1761		if (idx < s_idx)
1762			goto cont;
1763		if (idx > s_idx)
1764			s_q_idx = 0;
1765		q_idx = 0;
1766
1767		if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1768				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1769			goto done;
1770
1771		dev_queue = dev_ingress_queue(dev);
1772		if (dev_queue &&
1773		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1774				       &q_idx, s_q_idx, false,
1775				       tca[TCA_DUMP_INVISIBLE]) < 0)
1776			goto done;
1777
1778cont:
1779		idx++;
1780	}
1781
1782done:
1783	cb->args[0] = idx;
1784	cb->args[1] = q_idx;
1785
1786	return skb->len;
1787}
1788
1789
1790
1791/************************************************
1792 *	Traffic classes manipulation.		*
1793 ************************************************/
1794
1795static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1796			  unsigned long cl,
1797			  u32 portid, u32 seq, u16 flags, int event)
1798{
1799	struct tcmsg *tcm;
1800	struct nlmsghdr  *nlh;
1801	unsigned char *b = skb_tail_pointer(skb);
1802	struct gnet_dump d;
1803	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1804
1805	cond_resched();
1806	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1807	if (!nlh)
1808		goto out_nlmsg_trim;
1809	tcm = nlmsg_data(nlh);
1810	tcm->tcm_family = AF_UNSPEC;
1811	tcm->tcm__pad1 = 0;
1812	tcm->tcm__pad2 = 0;
1813	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1814	tcm->tcm_parent = q->handle;
1815	tcm->tcm_handle = q->handle;
1816	tcm->tcm_info = 0;
1817	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1818		goto nla_put_failure;
1819	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1820		goto nla_put_failure;
1821
1822	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1823					 NULL, &d, TCA_PAD) < 0)
1824		goto nla_put_failure;
1825
1826	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1827		goto nla_put_failure;
1828
1829	if (gnet_stats_finish_copy(&d) < 0)
1830		goto nla_put_failure;
1831
1832	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1833	return skb->len;
1834
1835out_nlmsg_trim:
1836nla_put_failure:
1837	nlmsg_trim(skb, b);
1838	return -1;
1839}
1840
1841static int tclass_notify(struct net *net, struct sk_buff *oskb,
1842			 struct nlmsghdr *n, struct Qdisc *q,
1843			 unsigned long cl, int event)
1844{
1845	struct sk_buff *skb;
1846	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1847	int err = 0;
1848
1849	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1850	if (!skb)
1851		return -ENOBUFS;
1852
1853	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1854		kfree_skb(skb);
1855		return -EINVAL;
1856	}
1857
1858	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1859			     n->nlmsg_flags & NLM_F_ECHO);
1860	if (err > 0)
1861		err = 0;
1862	return err;
1863}
1864
1865static int tclass_del_notify(struct net *net,
1866			     const struct Qdisc_class_ops *cops,
1867			     struct sk_buff *oskb, struct nlmsghdr *n,
1868			     struct Qdisc *q, unsigned long cl)
 
1869{
1870	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1871	struct sk_buff *skb;
1872	int err = 0;
1873
1874	if (!cops->delete)
1875		return -EOPNOTSUPP;
1876
1877	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1878	if (!skb)
1879		return -ENOBUFS;
1880
1881	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1882			   RTM_DELTCLASS) < 0) {
1883		kfree_skb(skb);
1884		return -EINVAL;
1885	}
1886
1887	err = cops->delete(q, cl);
1888	if (err) {
1889		kfree_skb(skb);
1890		return err;
1891	}
1892
1893	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1894			     n->nlmsg_flags & NLM_F_ECHO);
1895	if (err > 0)
1896		err = 0;
1897	return err;
1898}
1899
1900#ifdef CONFIG_NET_CLS
1901
1902struct tcf_bind_args {
1903	struct tcf_walker w;
1904	unsigned long base;
1905	unsigned long cl;
1906	u32 classid;
1907};
1908
1909static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1910{
1911	struct tcf_bind_args *a = (void *)arg;
1912
1913	if (tp->ops->bind_class) {
1914		struct Qdisc *q = tcf_block_q(tp->chain->block);
1915
1916		sch_tree_lock(q);
1917		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1918		sch_tree_unlock(q);
1919	}
1920	return 0;
1921}
1922
1923struct tc_bind_class_args {
1924	struct qdisc_walker w;
1925	unsigned long new_cl;
1926	u32 portid;
1927	u32 clid;
1928};
1929
1930static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1931				struct qdisc_walker *w)
1932{
1933	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1934	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1935	struct tcf_block *block;
1936	struct tcf_chain *chain;
1937
1938	block = cops->tcf_block(q, cl, NULL);
1939	if (!block)
1940		return 0;
1941	for (chain = tcf_get_next_chain(block, NULL);
1942	     chain;
1943	     chain = tcf_get_next_chain(block, chain)) {
1944		struct tcf_proto *tp;
1945
1946		for (tp = tcf_get_next_proto(chain, NULL, true);
1947		     tp; tp = tcf_get_next_proto(chain, tp, true)) {
1948			struct tcf_bind_args arg = {};
1949
1950			arg.w.fn = tcf_node_bind;
1951			arg.classid = a->clid;
1952			arg.base = cl;
1953			arg.cl = a->new_cl;
1954			tp->ops->walk(tp, &arg.w, true);
1955		}
1956	}
1957
1958	return 0;
1959}
1960
1961static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1962			   unsigned long new_cl)
1963{
1964	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1965	struct tc_bind_class_args args = {};
1966
1967	if (!cops->tcf_block)
1968		return;
1969	args.portid = portid;
1970	args.clid = clid;
1971	args.new_cl = new_cl;
1972	args.w.fn = tc_bind_class_walker;
1973	q->ops->cl_ops->walk(q, &args.w);
1974}
1975
1976#else
1977
1978static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1979			   unsigned long new_cl)
1980{
1981}
1982
1983#endif
1984
1985static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1986			 struct netlink_ext_ack *extack)
1987{
1988	struct net *net = sock_net(skb->sk);
1989	struct tcmsg *tcm = nlmsg_data(n);
1990	struct nlattr *tca[TCA_MAX + 1];
1991	struct net_device *dev;
1992	struct Qdisc *q = NULL;
1993	const struct Qdisc_class_ops *cops;
1994	unsigned long cl = 0;
1995	unsigned long new_cl;
1996	u32 portid;
1997	u32 clid;
1998	u32 qid;
1999	int err;
2000
2001	if ((n->nlmsg_type != RTM_GETTCLASS) &&
2002	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2003		return -EPERM;
2004
2005	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2006				     rtm_tca_policy, extack);
2007	if (err < 0)
2008		return err;
2009
2010	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2011	if (!dev)
2012		return -ENODEV;
2013
2014	/*
2015	   parent == TC_H_UNSPEC - unspecified parent.
2016	   parent == TC_H_ROOT   - class is root, which has no parent.
2017	   parent == X:0	 - parent is root class.
2018	   parent == X:Y	 - parent is a node in hierarchy.
2019	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2020
2021	   handle == 0:0	 - generate handle from kernel pool.
2022	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2023	   handle == X:Y	 - clear.
2024	   handle == X:0	 - root class.
2025	 */
2026
2027	/* Step 1. Determine qdisc handle X:0 */
2028
2029	portid = tcm->tcm_parent;
2030	clid = tcm->tcm_handle;
2031	qid = TC_H_MAJ(clid);
2032
2033	if (portid != TC_H_ROOT) {
2034		u32 qid1 = TC_H_MAJ(portid);
2035
2036		if (qid && qid1) {
2037			/* If both majors are known, they must be identical. */
2038			if (qid != qid1)
2039				return -EINVAL;
2040		} else if (qid1) {
2041			qid = qid1;
2042		} else if (qid == 0)
2043			qid = dev->qdisc->handle;
2044
2045		/* Now qid is genuine qdisc handle consistent
2046		 * both with parent and child.
2047		 *
2048		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2049		 */
2050		if (portid)
2051			portid = TC_H_MAKE(qid, portid);
2052	} else {
2053		if (qid == 0)
2054			qid = dev->qdisc->handle;
2055	}
2056
2057	/* OK. Locate qdisc */
2058	q = qdisc_lookup(dev, qid);
2059	if (!q)
2060		return -ENOENT;
2061
2062	/* An check that it supports classes */
2063	cops = q->ops->cl_ops;
2064	if (cops == NULL)
2065		return -EINVAL;
2066
2067	/* Now try to get class */
2068	if (clid == 0) {
2069		if (portid == TC_H_ROOT)
2070			clid = qid;
2071	} else
2072		clid = TC_H_MAKE(qid, clid);
2073
2074	if (clid)
2075		cl = cops->find(q, clid);
2076
2077	if (cl == 0) {
2078		err = -ENOENT;
2079		if (n->nlmsg_type != RTM_NEWTCLASS ||
2080		    !(n->nlmsg_flags & NLM_F_CREATE))
2081			goto out;
2082	} else {
2083		switch (n->nlmsg_type) {
2084		case RTM_NEWTCLASS:
2085			err = -EEXIST;
2086			if (n->nlmsg_flags & NLM_F_EXCL)
2087				goto out;
2088			break;
2089		case RTM_DELTCLASS:
2090			err = tclass_del_notify(net, cops, skb, n, q, cl);
2091			/* Unbind the class with flilters with 0 */
2092			tc_bind_tclass(q, portid, clid, 0);
2093			goto out;
2094		case RTM_GETTCLASS:
2095			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
2096			goto out;
2097		default:
2098			err = -EINVAL;
2099			goto out;
2100		}
2101	}
2102
2103	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2104		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2105		return -EOPNOTSUPP;
2106	}
2107
2108	new_cl = cl;
2109	err = -EOPNOTSUPP;
2110	if (cops->change)
2111		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2112	if (err == 0) {
2113		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
2114		/* We just create a new class, need to do reverse binding. */
2115		if (cl != new_cl)
2116			tc_bind_tclass(q, portid, clid, new_cl);
2117	}
2118out:
2119	return err;
2120}
2121
2122struct qdisc_dump_args {
2123	struct qdisc_walker	w;
2124	struct sk_buff		*skb;
2125	struct netlink_callback	*cb;
2126};
2127
2128static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2129			    struct qdisc_walker *arg)
2130{
2131	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2132
2133	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2134			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2135			      RTM_NEWTCLASS);
2136}
2137
2138static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2139				struct tcmsg *tcm, struct netlink_callback *cb,
2140				int *t_p, int s_t)
2141{
2142	struct qdisc_dump_args arg;
2143
2144	if (tc_qdisc_dump_ignore(q, false) ||
2145	    *t_p < s_t || !q->ops->cl_ops ||
2146	    (tcm->tcm_parent &&
2147	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2148		(*t_p)++;
2149		return 0;
2150	}
2151	if (*t_p > s_t)
2152		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2153	arg.w.fn = qdisc_class_dump;
2154	arg.skb = skb;
2155	arg.cb = cb;
2156	arg.w.stop  = 0;
2157	arg.w.skip = cb->args[1];
2158	arg.w.count = 0;
2159	q->ops->cl_ops->walk(q, &arg.w);
2160	cb->args[1] = arg.w.count;
2161	if (arg.w.stop)
2162		return -1;
2163	(*t_p)++;
2164	return 0;
2165}
2166
2167static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2168			       struct tcmsg *tcm, struct netlink_callback *cb,
2169			       int *t_p, int s_t)
2170{
2171	struct Qdisc *q;
2172	int b;
2173
2174	if (!root)
2175		return 0;
2176
2177	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2178		return -1;
2179
2180	if (!qdisc_dev(root))
2181		return 0;
2182
2183	if (tcm->tcm_parent) {
2184		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2185		if (q && q != root &&
2186		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2187			return -1;
2188		return 0;
2189	}
2190	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2191		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2192			return -1;
2193	}
2194
2195	return 0;
2196}
2197
2198static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2199{
2200	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2201	struct net *net = sock_net(skb->sk);
2202	struct netdev_queue *dev_queue;
2203	struct net_device *dev;
2204	int t, s_t;
2205
2206	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2207		return 0;
2208	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2209	if (!dev)
2210		return 0;
2211
2212	s_t = cb->args[0];
2213	t = 0;
2214
2215	if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
2216		goto done;
2217
2218	dev_queue = dev_ingress_queue(dev);
2219	if (dev_queue &&
2220	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2221				&t, s_t) < 0)
2222		goto done;
2223
2224done:
2225	cb->args[0] = t;
2226
2227	dev_put(dev);
2228	return skb->len;
2229}
2230
2231#ifdef CONFIG_PROC_FS
2232static int psched_show(struct seq_file *seq, void *v)
2233{
2234	seq_printf(seq, "%08x %08x %08x %08x\n",
2235		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2236		   1000000,
2237		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2238
2239	return 0;
2240}
2241
2242static int __net_init psched_net_init(struct net *net)
2243{
2244	struct proc_dir_entry *e;
2245
2246	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2247	if (e == NULL)
2248		return -ENOMEM;
2249
2250	return 0;
2251}
2252
2253static void __net_exit psched_net_exit(struct net *net)
2254{
2255	remove_proc_entry("psched", net->proc_net);
2256}
2257#else
2258static int __net_init psched_net_init(struct net *net)
2259{
2260	return 0;
2261}
2262
2263static void __net_exit psched_net_exit(struct net *net)
2264{
2265}
2266#endif
2267
2268static struct pernet_operations psched_net_ops = {
2269	.init = psched_net_init,
2270	.exit = psched_net_exit,
2271};
2272
2273static int __init pktsched_init(void)
2274{
2275	int err;
2276
2277	err = register_pernet_subsys(&psched_net_ops);
2278	if (err) {
2279		pr_err("pktsched_init: "
2280		       "cannot initialize per netns operations\n");
2281		return err;
2282	}
2283
2284	register_qdisc(&pfifo_fast_ops);
2285	register_qdisc(&pfifo_qdisc_ops);
2286	register_qdisc(&bfifo_qdisc_ops);
2287	register_qdisc(&pfifo_head_drop_qdisc_ops);
2288	register_qdisc(&mq_qdisc_ops);
2289	register_qdisc(&noqueue_qdisc_ops);
2290
2291	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2292	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2293	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2294		      0);
2295	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2296	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2297	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2298		      0);
2299
2300	return 0;
2301}
2302
2303subsys_initcall(pktsched_init);