cls_u32.c

   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * net/sched/cls_u32.c	Ugly (or Universal) 32bit key Packet Classifier.
   4 *
   5 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   6 *
   7 *	The filters are packed to hash tables of key nodes
   8 *	with a set of 32bit key/mask pairs at every node.
   9 *	Nodes reference next level hash tables etc.
  10 *
  11 *	This scheme is the best universal classifier I managed to
  12 *	invent; it is not super-fast, but it is not slow (provided you
  13 *	program it correctly), and general enough.  And its relative
  14 *	speed grows as the number of rules becomes larger.
  15 *
  16 *	It seems that it represents the best middle point between
  17 *	speed and manageability both by human and by machine.
  18 *
  19 *	It is especially useful for link sharing combined with QoS;
  20 *	pure RSVP doesn't need such a general approach and can use
  21 *	much simpler (and faster) schemes, sort of cls_rsvp.c.
  22 *
  23 *	nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
  24 */
  25
  26#include <linux/module.h>
  27#include <linux/slab.h>
  28#include <linux/types.h>
  29#include <linux/kernel.h>
  30#include <linux/string.h>
  31#include <linux/errno.h>
  32#include <linux/percpu.h>
  33#include <linux/rtnetlink.h>
  34#include <linux/skbuff.h>
  35#include <linux/bitmap.h>
  36#include <linux/netdevice.h>
  37#include <linux/hash.h>
  38#include <net/netlink.h>
  39#include <net/act_api.h>
  40#include <net/pkt_cls.h>
  41#include <linux/idr.h>
  42#include <net/tc_wrapper.h>
  43
  44struct tc_u_knode {
  45	struct tc_u_knode __rcu	*next;
  46	u32			handle;
  47	struct tc_u_hnode __rcu	*ht_up;
  48	struct tcf_exts		exts;
  49	int			ifindex;
  50	u8			fshift;
  51	struct tcf_result	res;
  52	struct tc_u_hnode __rcu	*ht_down;
  53#ifdef CONFIG_CLS_U32_PERF
  54	struct tc_u32_pcnt __percpu *pf;
  55#endif
  56	u32			flags;
  57	unsigned int		in_hw_count;
  58#ifdef CONFIG_CLS_U32_MARK
  59	u32			val;
  60	u32			mask;
  61	u32 __percpu		*pcpu_success;
  62#endif
  63	struct rcu_work		rwork;
  64	/* The 'sel' field MUST be the last field in structure to allow for
  65	 * tc_u32_keys allocated at end of structure.
  66	 */
  67	struct tc_u32_sel	sel;
  68};
  69
  70struct tc_u_hnode {
  71	struct tc_u_hnode __rcu	*next;
  72	u32			handle;
  73	u32			prio;
  74	refcount_t		refcnt;
  75	unsigned int		divisor;
  76	struct idr		handle_idr;
  77	bool			is_root;
  78	struct rcu_head		rcu;
  79	u32			flags;
  80	/* The 'ht' field MUST be the last field in structure to allow for
  81	 * more entries allocated at end of structure.
  82	 */
  83	struct tc_u_knode __rcu	*ht[];
  84};
  85
  86struct tc_u_common {
  87	struct tc_u_hnode __rcu	*hlist;
  88	void			*ptr;
  89	refcount_t		refcnt;
  90	struct idr		handle_idr;
  91	struct hlist_node	hnode;
  92	long			knodes;
  93};
  94
  95static u32 handle2id(u32 h)
  96{
  97	return ((h & 0x80000000) ? ((h >> 20) & 0x7FF) : h);
  98}
  99
 100static u32 id2handle(u32 id)
 101{
 102	return (id | 0x800U) << 20;
 103}
 104
 105static inline unsigned int u32_hash_fold(__be32 key,
 106					 const struct tc_u32_sel *sel,
 107					 u8 fshift)
 108{
 109	unsigned int h = ntohl(key & sel->hmask) >> fshift;
 110
 111	return h;
 112}
 113
 114TC_INDIRECT_SCOPE int u32_classify(struct sk_buff *skb,
 115				   const struct tcf_proto *tp,
 116				   struct tcf_result *res)
 117{
 118	struct {
 119		struct tc_u_knode *knode;
 120		unsigned int	  off;
 121	} stack[TC_U32_MAXDEPTH];
 122
 123	struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
 124	unsigned int off = skb_network_offset(skb);
 125	struct tc_u_knode *n;
 126	int sdepth = 0;
 127	int off2 = 0;
 128	int sel = 0;
 129#ifdef CONFIG_CLS_U32_PERF
 130	int j;
 131#endif
 132	int i, r;
 133
 134next_ht:
 135	n = rcu_dereference_bh(ht->ht[sel]);
 136
 137next_knode:
 138	if (n) {
 139		struct tc_u32_key *key = n->sel.keys;
 140
 141#ifdef CONFIG_CLS_U32_PERF
 142		__this_cpu_inc(n->pf->rcnt);
 143		j = 0;
 144#endif
 145
 146		if (tc_skip_sw(n->flags)) {
 147			n = rcu_dereference_bh(n->next);
 148			goto next_knode;
 149		}
 150
 151#ifdef CONFIG_CLS_U32_MARK
 152		if ((skb->mark & n->mask) != n->val) {
 153			n = rcu_dereference_bh(n->next);
 154			goto next_knode;
 155		} else {
 156			__this_cpu_inc(*n->pcpu_success);
 157		}
 158#endif
 159
 160		for (i = n->sel.nkeys; i > 0; i--, key++) {
 161			int toff = off + key->off + (off2 & key->offmask);
 162			__be32 *data, hdata;
 163
 164			data = skb_header_pointer_careful(skb, toff, 4,
 165							  &hdata);
 166			if (!data)
 167				goto out;
 168			if ((*data ^ key->val) & key->mask) {
 169				n = rcu_dereference_bh(n->next);
 170				goto next_knode;
 171			}
 172#ifdef CONFIG_CLS_U32_PERF
 173			__this_cpu_inc(n->pf->kcnts[j]);
 174			j++;
 175#endif
 176		}
 177
 178		ht = rcu_dereference_bh(n->ht_down);
 179		if (!ht) {
 180check_terminal:
 181			if (n->sel.flags & TC_U32_TERMINAL) {
 182
 183				*res = n->res;
 184				if (!tcf_match_indev(skb, n->ifindex)) {
 185					n = rcu_dereference_bh(n->next);
 186					goto next_knode;
 187				}
 188#ifdef CONFIG_CLS_U32_PERF
 189				__this_cpu_inc(n->pf->rhit);
 190#endif
 191				r = tcf_exts_exec(skb, &n->exts, res);
 192				if (r < 0) {
 193					n = rcu_dereference_bh(n->next);
 194					goto next_knode;
 195				}
 196
 197				return r;
 198			}
 199			n = rcu_dereference_bh(n->next);
 200			goto next_knode;
 201		}
 202
 203		/* PUSH */
 204		if (sdepth >= TC_U32_MAXDEPTH)
 205			goto deadloop;
 206		stack[sdepth].knode = n;
 207		stack[sdepth].off = off;
 208		sdepth++;
 209
 210		ht = rcu_dereference_bh(n->ht_down);
 211		sel = 0;
 212		if (ht->divisor) {
 213			__be32 *data, hdata;
 214
 215			data = skb_header_pointer_careful(skb,
 216							  off + n->sel.hoff,
 217							  4, &hdata);
 218			if (!data)
 219				goto out;
 220			sel = ht->divisor & u32_hash_fold(*data, &n->sel,
 221							  n->fshift);
 222		}
 223		if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
 224			goto next_ht;
 225
 226		if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
 227			off2 = n->sel.off + 3;
 228			if (n->sel.flags & TC_U32_VAROFFSET) {
 229				__be16 *data, hdata;
 230
 231				data = skb_header_pointer_careful(skb,
 232							  off + n->sel.offoff,
 233							  2, &hdata);
 234				if (!data)
 235					goto out;
 236				off2 += ntohs(n->sel.offmask & *data) >>
 237					n->sel.offshift;
 238			}
 239			off2 &= ~3;
 240		}
 241		if (n->sel.flags & TC_U32_EAT) {
 242			off += off2;
 243			off2 = 0;
 244		}
 245
 246		if (off < skb->len)
 247			goto next_ht;
 248	}
 249
 250	/* POP */
 251	if (sdepth--) {
 252		n = stack[sdepth].knode;
 253		ht = rcu_dereference_bh(n->ht_up);
 254		off = stack[sdepth].off;
 255		goto check_terminal;
 256	}
 257out:
 258	return -1;
 259
 260deadloop:
 261	net_warn_ratelimited("cls_u32: dead loop\n");
 262	return -1;
 263}
 264
 265static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
 266{
 267	struct tc_u_hnode *ht;
 268
 269	for (ht = rtnl_dereference(tp_c->hlist);
 270	     ht;
 271	     ht = rtnl_dereference(ht->next))
 272		if (ht->handle == handle)
 273			break;
 274
 275	return ht;
 276}
 277
 278static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
 279{
 280	unsigned int sel;
 281	struct tc_u_knode *n = NULL;
 282
 283	sel = TC_U32_HASH(handle);
 284	if (sel > ht->divisor)
 285		goto out;
 286
 287	for (n = rtnl_dereference(ht->ht[sel]);
 288	     n;
 289	     n = rtnl_dereference(n->next))
 290		if (n->handle == handle)
 291			break;
 292out:
 293	return n;
 294}
 295
 296
 297static void *u32_get(struct tcf_proto *tp, u32 handle)
 298{
 299	struct tc_u_hnode *ht;
 300	struct tc_u_common *tp_c = tp->data;
 301
 302	if (TC_U32_HTID(handle) == TC_U32_ROOT)
 303		ht = rtnl_dereference(tp->root);
 304	else
 305		ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
 306
 307	if (!ht)
 308		return NULL;
 309
 310	if (TC_U32_KEY(handle) == 0)
 311		return ht;
 312
 313	return u32_lookup_key(ht, handle);
 314}
 315
 316/* Protected by rtnl lock */
 317static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
 318{
 319	int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
 320	if (id < 0)
 321		return 0;
 322	return id2handle(id);
 323}
 324
 325static struct hlist_head *tc_u_common_hash;
 326
 327#define U32_HASH_SHIFT 10
 328#define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
 329
 330static void *tc_u_common_ptr(const struct tcf_proto *tp)
 331{
 332	struct tcf_block *block = tp->chain->block;
 333
 334	/* The block sharing is currently supported only
 335	 * for classless qdiscs. In that case we use block
 336	 * for tc_u_common identification. In case the
 337	 * block is not shared, block->q is a valid pointer
 338	 * and we can use that. That works for classful qdiscs.
 339	 */
 340	if (tcf_block_shared(block))
 341		return block;
 342	else
 343		return block->q;
 344}
 345
 346static struct hlist_head *tc_u_hash(void *key)
 347{
 348	return tc_u_common_hash + hash_ptr(key, U32_HASH_SHIFT);
 349}
 350
 351static struct tc_u_common *tc_u_common_find(void *key)
 352{
 353	struct tc_u_common *tc;
 354	hlist_for_each_entry(tc, tc_u_hash(key), hnode) {
 355		if (tc->ptr == key)
 356			return tc;
 357	}
 358	return NULL;
 359}
 360
 361static int u32_init(struct tcf_proto *tp)
 362{
 363	struct tc_u_hnode *root_ht;
 364	void *key = tc_u_common_ptr(tp);
 365	struct tc_u_common *tp_c = tc_u_common_find(key);
 366
 367	root_ht = kzalloc(struct_size(root_ht, ht, 1), GFP_KERNEL);
 368	if (root_ht == NULL)
 369		return -ENOBUFS;
 370
 371	refcount_set(&root_ht->refcnt, 1);
 372	root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : id2handle(0);
 373	root_ht->prio = tp->prio;
 374	root_ht->is_root = true;
 375	idr_init(&root_ht->handle_idr);
 376
 377	if (tp_c == NULL) {
 378		tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
 379		if (tp_c == NULL) {
 380			kfree(root_ht);
 381			return -ENOBUFS;
 382		}
 383		refcount_set(&tp_c->refcnt, 1);
 384		tp_c->ptr = key;
 385		INIT_HLIST_NODE(&tp_c->hnode);
 386		idr_init(&tp_c->handle_idr);
 387
 388		hlist_add_head(&tp_c->hnode, tc_u_hash(key));
 389	} else {
 390		refcount_inc(&tp_c->refcnt);
 391	}
 392
 393	RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
 394	rcu_assign_pointer(tp_c->hlist, root_ht);
 395
 396	/* root_ht must be destroyed when tcf_proto is destroyed */
 397	rcu_assign_pointer(tp->root, root_ht);
 398	tp->data = tp_c;
 399	return 0;
 400}
 401
 402static void __u32_destroy_key(struct tc_u_knode *n)
 403{
 404	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 405
 406	tcf_exts_destroy(&n->exts);
 407	if (ht && refcount_dec_and_test(&ht->refcnt))
 408		kfree(ht);
 409	kfree(n);
 410}
 411
 412static void u32_destroy_key(struct tc_u_knode *n, bool free_pf)
 413{
 414	tcf_exts_put_net(&n->exts);
 415#ifdef CONFIG_CLS_U32_PERF
 416	if (free_pf)
 417		free_percpu(n->pf);
 418#endif
 419#ifdef CONFIG_CLS_U32_MARK
 420	if (free_pf)
 421		free_percpu(n->pcpu_success);
 422#endif
 423	__u32_destroy_key(n);
 424}
 425
 426/* u32_delete_key_rcu should be called when free'ing a copied
 427 * version of a tc_u_knode obtained from u32_init_knode(). When
 428 * copies are obtained from u32_init_knode() the statistics are
 429 * shared between the old and new copies to allow readers to
 430 * continue to update the statistics during the copy. To support
 431 * this the u32_delete_key_rcu variant does not free the percpu
 432 * statistics.
 433 */
 434static void u32_delete_key_work(struct work_struct *work)
 435{
 436	struct tc_u_knode *key = container_of(to_rcu_work(work),
 437					      struct tc_u_knode,
 438					      rwork);
 439	rtnl_lock();
 440	u32_destroy_key(key, false);
 441	rtnl_unlock();
 442}
 443
 444/* u32_delete_key_freepf_rcu is the rcu callback variant
 445 * that free's the entire structure including the statistics
 446 * percpu variables. Only use this if the key is not a copy
 447 * returned by u32_init_knode(). See u32_delete_key_rcu()
 448 * for the variant that should be used with keys return from
 449 * u32_init_knode()
 450 */
 451static void u32_delete_key_freepf_work(struct work_struct *work)
 452{
 453	struct tc_u_knode *key = container_of(to_rcu_work(work),
 454					      struct tc_u_knode,
 455					      rwork);
 456	rtnl_lock();
 457	u32_destroy_key(key, true);
 458	rtnl_unlock();
 459}
 460
 461static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
 462{
 463	struct tc_u_common *tp_c = tp->data;
 464	struct tc_u_knode __rcu **kp;
 465	struct tc_u_knode *pkp;
 466	struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
 467
 468	if (ht) {
 469		kp = &ht->ht[TC_U32_HASH(key->handle)];
 470		for (pkp = rtnl_dereference(*kp); pkp;
 471		     kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
 472			if (pkp == key) {
 473				RCU_INIT_POINTER(*kp, key->next);
 474				tp_c->knodes--;
 475
 476				tcf_unbind_filter(tp, &key->res);
 477				idr_remove(&ht->handle_idr, key->handle);
 478				tcf_exts_get_net(&key->exts);
 479				tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
 480				return 0;
 481			}
 482		}
 483	}
 484	WARN_ON(1);
 485	return 0;
 486}
 487
 488static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 489			       struct netlink_ext_ack *extack)
 490{
 491	struct tcf_block *block = tp->chain->block;
 492	struct tc_cls_u32_offload cls_u32 = {};
 493
 494	tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
 495	cls_u32.command = TC_CLSU32_DELETE_HNODE;
 496	cls_u32.hnode.divisor = h->divisor;
 497	cls_u32.hnode.handle = h->handle;
 498	cls_u32.hnode.prio = h->prio;
 499
 500	tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false, true);
 501}
 502
 503static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 504				u32 flags, struct netlink_ext_ack *extack)
 505{
 506	struct tcf_block *block = tp->chain->block;
 507	struct tc_cls_u32_offload cls_u32 = {};
 508	bool skip_sw = tc_skip_sw(flags);
 509	bool offloaded = false;
 510	int err;
 511
 512	tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 513	cls_u32.command = TC_CLSU32_NEW_HNODE;
 514	cls_u32.hnode.divisor = h->divisor;
 515	cls_u32.hnode.handle = h->handle;
 516	cls_u32.hnode.prio = h->prio;
 517
 518	err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw, true);
 519	if (err < 0) {
 520		u32_clear_hw_hnode(tp, h, NULL);
 521		return err;
 522	} else if (err > 0) {
 523		offloaded = true;
 524	}
 525
 526	if (skip_sw && !offloaded)
 527		return -EINVAL;
 528
 529	return 0;
 530}
 531
 532static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 533				struct netlink_ext_ack *extack)
 534{
 535	struct tcf_block *block = tp->chain->block;
 536	struct tc_cls_u32_offload cls_u32 = {};
 537
 538	tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
 539	cls_u32.command = TC_CLSU32_DELETE_KNODE;
 540	cls_u32.knode.handle = n->handle;
 541
 542	tc_setup_cb_destroy(block, tp, TC_SETUP_CLSU32, &cls_u32, false,
 543			    &n->flags, &n->in_hw_count, true);
 544}
 545
 546static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 547				u32 flags, struct netlink_ext_ack *extack)
 548{
 549	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 550	struct tcf_block *block = tp->chain->block;
 551	struct tc_cls_u32_offload cls_u32 = {};
 552	bool skip_sw = tc_skip_sw(flags);
 553	int err;
 554
 555	tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 556	cls_u32.command = TC_CLSU32_REPLACE_KNODE;
 557	cls_u32.knode.handle = n->handle;
 558	cls_u32.knode.fshift = n->fshift;
 559#ifdef CONFIG_CLS_U32_MARK
 560	cls_u32.knode.val = n->val;
 561	cls_u32.knode.mask = n->mask;
 562#else
 563	cls_u32.knode.val = 0;
 564	cls_u32.knode.mask = 0;
 565#endif
 566	cls_u32.knode.sel = &n->sel;
 567	cls_u32.knode.res = &n->res;
 568	cls_u32.knode.exts = &n->exts;
 569	if (n->ht_down)
 570		cls_u32.knode.link_handle = ht->handle;
 571
 572	err = tc_setup_cb_add(block, tp, TC_SETUP_CLSU32, &cls_u32, skip_sw,
 573			      &n->flags, &n->in_hw_count, true);
 574	if (err) {
 575		u32_remove_hw_knode(tp, n, NULL);
 576		return err;
 577	}
 578
 579	if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
 580		return -EINVAL;
 581
 582	return 0;
 583}
 584
 585static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 586			    struct netlink_ext_ack *extack)
 587{
 588	struct tc_u_common *tp_c = tp->data;
 589	struct tc_u_knode *n;
 590	unsigned int h;
 591
 592	for (h = 0; h <= ht->divisor; h++) {
 593		while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
 594			RCU_INIT_POINTER(ht->ht[h],
 595					 rtnl_dereference(n->next));
 596			tp_c->knodes--;
 597			tcf_unbind_filter(tp, &n->res);
 598			u32_remove_hw_knode(tp, n, extack);
 599			idr_remove(&ht->handle_idr, n->handle);
 600			if (tcf_exts_get_net(&n->exts))
 601				tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
 602			else
 603				u32_destroy_key(n, true);
 604		}
 605	}
 606}
 607
 608static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 609			     struct netlink_ext_ack *extack)
 610{
 611	struct tc_u_common *tp_c = tp->data;
 612	struct tc_u_hnode __rcu **hn;
 613	struct tc_u_hnode *phn;
 614
 615	u32_clear_hnode(tp, ht, extack);
 616
 617	hn = &tp_c->hlist;
 618	for (phn = rtnl_dereference(*hn);
 619	     phn;
 620	     hn = &phn->next, phn = rtnl_dereference(*hn)) {
 621		if (phn == ht) {
 622			u32_clear_hw_hnode(tp, ht, extack);
 623			idr_destroy(&ht->handle_idr);
 624			idr_remove(&tp_c->handle_idr, handle2id(ht->handle));
 625			RCU_INIT_POINTER(*hn, ht->next);
 626			kfree_rcu(ht, rcu);
 627			return 0;
 628		}
 629	}
 630
 631	return -ENOENT;
 632}
 633
 634static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
 635			struct netlink_ext_ack *extack)
 636{
 637	struct tc_u_common *tp_c = tp->data;
 638	struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
 639
 640	WARN_ON(root_ht == NULL);
 641
 642	if (root_ht && refcount_dec_and_test(&root_ht->refcnt))
 643		u32_destroy_hnode(tp, root_ht, extack);
 644
 645	if (refcount_dec_and_test(&tp_c->refcnt)) {
 646		struct tc_u_hnode *ht;
 647
 648		hlist_del(&tp_c->hnode);
 649
 650		while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
 651			u32_clear_hnode(tp, ht, extack);
 652			RCU_INIT_POINTER(tp_c->hlist, ht->next);
 653
 654			/* u32_destroy_key() will later free ht for us, if it's
 655			 * still referenced by some knode
 656			 */
 657			if (refcount_dec_and_test(&ht->refcnt))
 658				kfree_rcu(ht, rcu);
 659		}
 660
 661		idr_destroy(&tp_c->handle_idr);
 662		kfree(tp_c);
 663	}
 664
 665	tp->data = NULL;
 666}
 667
 668static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
 669		      bool rtnl_held, struct netlink_ext_ack *extack)
 670{
 671	struct tc_u_hnode *ht = arg;
 672	struct tc_u_common *tp_c = tp->data;
 673	int ret = 0;
 674
 675	if (TC_U32_KEY(ht->handle)) {
 676		u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
 677		ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
 678		goto out;
 679	}
 680
 681	if (ht->is_root) {
 682		NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
 683		return -EINVAL;
 684	}
 685
 686	if (refcount_dec_if_one(&ht->refcnt)) {
 687		u32_destroy_hnode(tp, ht, extack);
 688	} else {
 689		NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
 690		return -EBUSY;
 691	}
 692
 693out:
 694	*last = refcount_read(&tp_c->refcnt) == 1 && tp_c->knodes == 0;
 695	return ret;
 696}
 697
 698static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
 699{
 700	u32 index = htid | 0x800;
 701	u32 max = htid | 0xFFF;
 702
 703	if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
 704		index = htid + 1;
 705		if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
 706				 GFP_KERNEL))
 707			index = max;
 708	}
 709
 710	return index;
 711}
 712
 713static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
 714	[TCA_U32_CLASSID]	= { .type = NLA_U32 },
 715	[TCA_U32_HASH]		= { .type = NLA_U32 },
 716	[TCA_U32_LINK]		= { .type = NLA_U32 },
 717	[TCA_U32_DIVISOR]	= { .type = NLA_U32 },
 718	[TCA_U32_SEL]		= { .len = sizeof(struct tc_u32_sel) },
 719	[TCA_U32_INDEV]		= { .type = NLA_STRING, .len = IFNAMSIZ },
 720	[TCA_U32_MARK]		= { .len = sizeof(struct tc_u32_mark) },
 721	[TCA_U32_FLAGS]		= { .type = NLA_U32 },
 722};
 723
 724static void u32_unbind_filter(struct tcf_proto *tp, struct tc_u_knode *n,
 725			      struct nlattr **tb)
 726{
 727	if (tb[TCA_U32_CLASSID])
 728		tcf_unbind_filter(tp, &n->res);
 729}
 730
 731static void u32_bind_filter(struct tcf_proto *tp, struct tc_u_knode *n,
 732			    unsigned long base, struct nlattr **tb)
 733{
 734	if (tb[TCA_U32_CLASSID]) {
 735		n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
 736		tcf_bind_filter(tp, &n->res, base);
 737	}
 738}
 739
 740static int u32_set_parms(struct net *net, struct tcf_proto *tp,
 741			 struct tc_u_knode *n, struct nlattr **tb,
 742			 struct nlattr *est, u32 flags, u32 fl_flags,
 743			 struct netlink_ext_ack *extack)
 744{
 745	int err, ifindex = -1;
 746
 747	err = tcf_exts_validate_ex(net, tp, tb, est, &n->exts, flags,
 748				   fl_flags, extack);
 749	if (err < 0)
 750		return err;
 751
 752	if (tb[TCA_U32_INDEV]) {
 753		ifindex = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
 754		if (ifindex < 0)
 755			return -EINVAL;
 756	}
 757
 758	if (tb[TCA_U32_LINK]) {
 759		u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
 760		struct tc_u_hnode *ht_down = NULL, *ht_old;
 761
 762		if (TC_U32_KEY(handle)) {
 763			NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
 764			return -EINVAL;
 765		}
 766
 767		if (handle) {
 768			ht_down = u32_lookup_ht(tp->data, handle);
 769
 770			if (!ht_down) {
 771				NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
 772				return -EINVAL;
 773			}
 774			if (ht_down->is_root) {
 775				NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
 776				return -EINVAL;
 777			}
 778			refcount_inc(&ht_down->refcnt);
 779		}
 780
 781		ht_old = rtnl_dereference(n->ht_down);
 782		rcu_assign_pointer(n->ht_down, ht_down);
 783
 784		if (ht_old)
 785			refcount_dec(&ht_old->refcnt);
 786	}
 787
 788	if (ifindex >= 0)
 789		n->ifindex = ifindex;
 790
 791	return 0;
 792}
 793
 794static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
 795			      struct tc_u_knode *n)
 796{
 797	struct tc_u_knode __rcu **ins;
 798	struct tc_u_knode *pins;
 799	struct tc_u_hnode *ht;
 800
 801	if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
 802		ht = rtnl_dereference(tp->root);
 803	else
 804		ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
 805
 806	ins = &ht->ht[TC_U32_HASH(n->handle)];
 807
 808	/* The node must always exist for it to be replaced if this is not the
 809	 * case then something went very wrong elsewhere.
 810	 */
 811	for (pins = rtnl_dereference(*ins); ;
 812	     ins = &pins->next, pins = rtnl_dereference(*ins))
 813		if (pins->handle == n->handle)
 814			break;
 815
 816	idr_replace(&ht->handle_idr, n, n->handle);
 817	RCU_INIT_POINTER(n->next, pins->next);
 818	rcu_assign_pointer(*ins, n);
 819}
 820
 821static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
 822					 struct tc_u_knode *n)
 823{
 824	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 825	struct tc_u32_sel *s = &n->sel;
 826	struct tc_u_knode *new;
 827
 828	new = kzalloc(struct_size(new, sel.keys, s->nkeys), GFP_KERNEL);
 829	if (!new)
 830		return NULL;
 831
 832	RCU_INIT_POINTER(new->next, n->next);
 833	new->handle = n->handle;
 834	RCU_INIT_POINTER(new->ht_up, n->ht_up);
 835
 836	new->ifindex = n->ifindex;
 837	new->fshift = n->fshift;
 838	new->flags = n->flags;
 839	RCU_INIT_POINTER(new->ht_down, ht);
 840
 841#ifdef CONFIG_CLS_U32_PERF
 842	/* Statistics may be incremented by readers during update
 843	 * so we must keep them in tact. When the node is later destroyed
 844	 * a special destroy call must be made to not free the pf memory.
 845	 */
 846	new->pf = n->pf;
 847#endif
 848
 849#ifdef CONFIG_CLS_U32_MARK
 850	new->val = n->val;
 851	new->mask = n->mask;
 852	/* Similarly success statistics must be moved as pointers */
 853	new->pcpu_success = n->pcpu_success;
 854#endif
 855	memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
 856
 857	if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
 858		kfree(new);
 859		return NULL;
 860	}
 861
 862	/* bump reference count as long as we hold pointer to structure */
 863	if (ht)
 864		refcount_inc(&ht->refcnt);
 865
 866	return new;
 867}
 868
 869static int u32_change(struct net *net, struct sk_buff *in_skb,
 870		      struct tcf_proto *tp, unsigned long base, u32 handle,
 871		      struct nlattr **tca, void **arg, u32 flags,
 872		      struct netlink_ext_ack *extack)
 873{
 874	struct tc_u_common *tp_c = tp->data;
 875	struct tc_u_hnode *ht;
 876	struct tc_u_knode *n;
 877	struct tc_u32_sel *s;
 878	struct nlattr *opt = tca[TCA_OPTIONS];
 879	struct nlattr *tb[TCA_U32_MAX + 1];
 880	u32 htid, userflags = 0;
 881	size_t sel_size;
 882	int err;
 883
 884	if (!opt) {
 885		if (handle) {
 886			NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
 887			return -EINVAL;
 888		} else {
 889			return 0;
 890		}
 891	}
 892
 893	err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
 894					  extack);
 895	if (err < 0)
 896		return err;
 897
 898	if (tb[TCA_U32_FLAGS]) {
 899		userflags = nla_get_u32(tb[TCA_U32_FLAGS]);
 900		if (!tc_flags_valid(userflags)) {
 901			NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
 902			return -EINVAL;
 903		}
 904	}
 905
 906	n = *arg;
 907	if (n) {
 908		struct tc_u_knode *new;
 909
 910		if (TC_U32_KEY(n->handle) == 0) {
 911			NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
 912			return -EINVAL;
 913		}
 914
 915		if ((n->flags ^ userflags) &
 916		    ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
 917			NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
 918			return -EINVAL;
 919		}
 920
 921		new = u32_init_knode(net, tp, n);
 922		if (!new)
 923			return -ENOMEM;
 924
 925		err = u32_set_parms(net, tp, new, tb, tca[TCA_RATE],
 926				    flags, new->flags, extack);
 927
 928		if (err) {
 929			__u32_destroy_key(new);
 930			return err;
 931		}
 932
 933		u32_bind_filter(tp, new, base, tb);
 934
 935		err = u32_replace_hw_knode(tp, new, flags, extack);
 936		if (err) {
 937			u32_unbind_filter(tp, new, tb);
 938
 939			if (tb[TCA_U32_LINK]) {
 940				struct tc_u_hnode *ht_old;
 941
 942				ht_old = rtnl_dereference(n->ht_down);
 943				if (ht_old)
 944					refcount_inc(&ht_old->refcnt);
 945			}
 946			__u32_destroy_key(new);
 947			return err;
 948		}
 949
 950		if (!tc_in_hw(new->flags))
 951			new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
 952
 953		tcf_proto_update_usesw(tp, new->flags);
 954
 955		u32_replace_knode(tp, tp_c, new);
 956		tcf_unbind_filter(tp, &n->res);
 957		tcf_exts_get_net(&n->exts);
 958		tcf_queue_work(&n->rwork, u32_delete_key_work);
 959		return 0;
 960	}
 961
 962	if (tb[TCA_U32_DIVISOR]) {
 963		unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
 964
 965		if (!is_power_of_2(divisor)) {
 966			NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
 967			return -EINVAL;
 968		}
 969		if (divisor-- > 0x100) {
 970			NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
 971			return -EINVAL;
 972		}
 973		if (TC_U32_KEY(handle)) {
 974			NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
 975			return -EINVAL;
 976		}
 977		ht = kzalloc(struct_size(ht, ht, divisor + 1), GFP_KERNEL);
 978		if (ht == NULL)
 979			return -ENOBUFS;
 980		if (handle == 0) {
 981			handle = gen_new_htid(tp->data, ht);
 982			if (handle == 0) {
 983				kfree(ht);
 984				return -ENOMEM;
 985			}
 986		} else {
 987			err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
 988					    handle, GFP_KERNEL);
 989			if (err) {
 990				kfree(ht);
 991				return err;
 992			}
 993		}
 994		refcount_set(&ht->refcnt, 1);
 995		ht->divisor = divisor;
 996		ht->handle = handle;
 997		ht->prio = tp->prio;
 998		idr_init(&ht->handle_idr);
 999		ht->flags = userflags;
1000
1001		err = u32_replace_hw_hnode(tp, ht, userflags, extack);
1002		if (err) {
1003			idr_remove(&tp_c->handle_idr, handle2id(handle));
1004			kfree(ht);
1005			return err;
1006		}
1007
1008		RCU_INIT_POINTER(ht->next, tp_c->hlist);
1009		rcu_assign_pointer(tp_c->hlist, ht);
1010		*arg = ht;
1011
1012		return 0;
1013	}
1014
1015	if (tb[TCA_U32_HASH]) {
1016		htid = nla_get_u32(tb[TCA_U32_HASH]);
1017		if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1018			ht = rtnl_dereference(tp->root);
1019			htid = ht->handle;
1020		} else {
1021			ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1022			if (!ht) {
1023				NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
1024				return -EINVAL;
1025			}
1026		}
1027	} else {
1028		ht = rtnl_dereference(tp->root);
1029		htid = ht->handle;
1030	}
1031
1032	if (ht->divisor < TC_U32_HASH(htid)) {
1033		NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1034		return -EINVAL;
1035	}
1036
1037	/* At this point, we need to derive the new handle that will be used to
1038	 * uniquely map the identity of this table match entry. The
1039	 * identity of the entry that we need to construct is 32 bits made of:
1040	 *     htid(12b):bucketid(8b):node/entryid(12b)
1041	 *
1042	 * At this point _we have the table(ht)_ in which we will insert this
1043	 * entry. We carry the table's id in variable "htid".
1044	 * Note that earlier code picked the ht selection either by a) the user
1045	 * providing the htid specified via TCA_U32_HASH attribute or b) when
1046	 * no such attribute is passed then the root ht, is default to at ID
1047	 * 0x[800][00][000]. Rule: the root table has a single bucket with ID 0.
1048	 * If OTOH the user passed us the htid, they may also pass a bucketid of
1049	 * choice. 0 is fine. For example a user htid is 0x[600][01][000] it is
1050	 * indicating hash bucketid of 1. Rule: the entry/node ID _cannot_ be
1051	 * passed via the htid, so even if it was non-zero it will be ignored.
1052	 *
1053	 * We may also have a handle, if the user passed one. The handle also
1054	 * carries the same addressing of htid(12b):bucketid(8b):node/entryid(12b).
1055	 * Rule: the bucketid on the handle is ignored even if one was passed;
1056	 * rather the value on "htid" is always assumed to be the bucketid.
1057	 */
1058	if (handle) {
1059		/* Rule: The htid from handle and tableid from htid must match */
1060		if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1061			NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1062			return -EINVAL;
1063		}
1064		/* Ok, so far we have a valid htid(12b):bucketid(8b) but we
1065		 * need to finalize the table entry identification with the last
1066		 * part - the node/entryid(12b)). Rule: Nodeid _cannot be 0_ for
1067		 * entries. Rule: nodeid of 0 is reserved only for tables(see
1068		 * earlier code which processes TC_U32_DIVISOR attribute).
1069		 * Rule: The nodeid can only be derived from the handle (and not
1070		 * htid).
1071		 * Rule: if the handle specified zero for the node id example
1072		 * 0x60000000, then pick a new nodeid from the pool of IDs
1073		 * this hash table has been allocating from.
1074		 * If OTOH it is specified (i.e for example the user passed a
1075		 * handle such as 0x60000123), then we use it generate our final
1076		 * handle which is used to uniquely identify the match entry.
1077		 */
1078		if (!TC_U32_NODE(handle)) {
1079			handle = gen_new_kid(ht, htid);
1080		} else {
1081			handle = htid | TC_U32_NODE(handle);
1082			err = idr_alloc_u32(&ht->handle_idr, NULL, &handle,
1083					    handle, GFP_KERNEL);
1084			if (err)
1085				return err;
1086		}
1087	} else {
1088		/* The user did not give us a handle; lets just generate one
1089		 * from the table's pool of nodeids.
1090		 */
1091		handle = gen_new_kid(ht, htid);
1092	}
1093
1094	if (tb[TCA_U32_SEL] == NULL) {
1095		NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1096		err = -EINVAL;
1097		goto erridr;
1098	}
1099
1100	s = nla_data(tb[TCA_U32_SEL]);
1101	sel_size = struct_size(s, keys, s->nkeys);
1102	if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1103		err = -EINVAL;
1104		goto erridr;
1105	}
1106
1107	n = kzalloc(struct_size(n, sel.keys, s->nkeys), GFP_KERNEL);
1108	if (n == NULL) {
1109		err = -ENOBUFS;
1110		goto erridr;
1111	}
1112
1113#ifdef CONFIG_CLS_U32_PERF
1114	n->pf = __alloc_percpu(struct_size(n->pf, kcnts, s->nkeys),
1115			       __alignof__(struct tc_u32_pcnt));
1116	if (!n->pf) {
1117		err = -ENOBUFS;
1118		goto errfree;
1119	}
1120#endif
1121
1122	unsafe_memcpy(&n->sel, s, sel_size,
1123		      /* A composite flex-array structure destination,
1124		       * which was correctly sized with struct_size(),
1125		       * bounds-checked against nla_len(), and allocated
1126		       * above. */);
1127	RCU_INIT_POINTER(n->ht_up, ht);
1128	n->handle = handle;
1129	n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1130	n->flags = userflags;
1131
1132	err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1133	if (err < 0)
1134		goto errout;
1135
1136#ifdef CONFIG_CLS_U32_MARK
1137	n->pcpu_success = alloc_percpu(u32);
1138	if (!n->pcpu_success) {
1139		err = -ENOMEM;
1140		goto errout;
1141	}
1142
1143	if (tb[TCA_U32_MARK]) {
1144		struct tc_u32_mark *mark;
1145
1146		mark = nla_data(tb[TCA_U32_MARK]);
1147		n->val = mark->val;
1148		n->mask = mark->mask;
1149	}
1150#endif
1151
1152	err = u32_set_parms(net, tp, n, tb, tca[TCA_RATE],
1153			    flags, n->flags, extack);
1154
1155	u32_bind_filter(tp, n, base, tb);
1156
1157	if (err == 0) {
1158		struct tc_u_knode __rcu **ins;
1159		struct tc_u_knode *pins;
1160
1161		err = u32_replace_hw_knode(tp, n, flags, extack);
1162		if (err)
1163			goto errunbind;
1164
1165		if (!tc_in_hw(n->flags))
1166			n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1167
1168		tcf_proto_update_usesw(tp, n->flags);
1169
1170		ins = &ht->ht[TC_U32_HASH(handle)];
1171		for (pins = rtnl_dereference(*ins); pins;
1172		     ins = &pins->next, pins = rtnl_dereference(*ins))
1173			if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1174				break;
1175
1176		RCU_INIT_POINTER(n->next, pins);
1177		rcu_assign_pointer(*ins, n);
1178		tp_c->knodes++;
1179		*arg = n;
1180		return 0;
1181	}
1182
1183errunbind:
1184	u32_unbind_filter(tp, n, tb);
1185
1186#ifdef CONFIG_CLS_U32_MARK
1187	free_percpu(n->pcpu_success);
1188#endif
1189
1190errout:
1191	tcf_exts_destroy(&n->exts);
1192#ifdef CONFIG_CLS_U32_PERF
1193errfree:
1194	free_percpu(n->pf);
1195#endif
1196	kfree(n);
1197erridr:
1198	idr_remove(&ht->handle_idr, handle);
1199	return err;
1200}
1201
1202static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1203		     bool rtnl_held)
1204{
1205	struct tc_u_common *tp_c = tp->data;
1206	struct tc_u_hnode *ht;
1207	struct tc_u_knode *n;
1208	unsigned int h;
1209
1210	if (arg->stop)
1211		return;
1212
1213	for (ht = rtnl_dereference(tp_c->hlist);
1214	     ht;
1215	     ht = rtnl_dereference(ht->next)) {
1216		if (ht->prio != tp->prio)
1217			continue;
1218
1219		if (!tc_cls_stats_dump(tp, arg, ht))
1220			return;
1221
1222		for (h = 0; h <= ht->divisor; h++) {
1223			for (n = rtnl_dereference(ht->ht[h]);
1224			     n;
1225			     n = rtnl_dereference(n->next)) {
1226				if (!tc_cls_stats_dump(tp, arg, n))
1227					return;
1228			}
1229		}
1230	}
1231}
1232
1233static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1234			       bool add, flow_setup_cb_t *cb, void *cb_priv,
1235			       struct netlink_ext_ack *extack)
1236{
1237	struct tc_cls_u32_offload cls_u32 = {};
1238	int err;
1239
1240	tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1241	cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1242	cls_u32.hnode.divisor = ht->divisor;
1243	cls_u32.hnode.handle = ht->handle;
1244	cls_u32.hnode.prio = ht->prio;
1245
1246	err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1247	if (err && add && tc_skip_sw(ht->flags))
1248		return err;
1249
1250	return 0;
1251}
1252
1253static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1254			       bool add, flow_setup_cb_t *cb, void *cb_priv,
1255			       struct netlink_ext_ack *extack)
1256{
1257	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1258	struct tcf_block *block = tp->chain->block;
1259	struct tc_cls_u32_offload cls_u32 = {};
1260
1261	tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1262	cls_u32.command = add ?
1263		TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1264	cls_u32.knode.handle = n->handle;
1265
1266	if (add) {
1267		cls_u32.knode.fshift = n->fshift;
1268#ifdef CONFIG_CLS_U32_MARK
1269		cls_u32.knode.val = n->val;
1270		cls_u32.knode.mask = n->mask;
1271#else
1272		cls_u32.knode.val = 0;
1273		cls_u32.knode.mask = 0;
1274#endif
1275		cls_u32.knode.sel = &n->sel;
1276		cls_u32.knode.res = &n->res;
1277		cls_u32.knode.exts = &n->exts;
1278		if (n->ht_down)
1279			cls_u32.knode.link_handle = ht->handle;
1280	}
1281
1282	return tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
1283				     &cls_u32, cb_priv, &n->flags,
1284				     &n->in_hw_count);
1285}
1286
1287static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1288			 void *cb_priv, struct netlink_ext_ack *extack)
1289{
1290	struct tc_u_common *tp_c = tp->data;
1291	struct tc_u_hnode *ht;
1292	struct tc_u_knode *n;
1293	unsigned int h;
1294	int err;
1295
1296	for (ht = rtnl_dereference(tp_c->hlist);
1297	     ht;
1298	     ht = rtnl_dereference(ht->next)) {
1299		if (ht->prio != tp->prio)
1300			continue;
1301
1302		/* When adding filters to a new dev, try to offload the
1303		 * hashtable first. When removing, do the filters before the
1304		 * hashtable.
1305		 */
1306		if (add && !tc_skip_hw(ht->flags)) {
1307			err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1308						  extack);
1309			if (err)
1310				return err;
1311		}
1312
1313		for (h = 0; h <= ht->divisor; h++) {
1314			for (n = rtnl_dereference(ht->ht[h]);
1315			     n;
1316			     n = rtnl_dereference(n->next)) {
1317				if (tc_skip_hw(n->flags))
1318					continue;
1319
1320				err = u32_reoffload_knode(tp, n, add, cb,
1321							  cb_priv, extack);
1322				if (err)
1323					return err;
1324			}
1325		}
1326
1327		if (!add && !tc_skip_hw(ht->flags))
1328			u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1329	}
1330
1331	return 0;
1332}
1333
1334static void u32_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
1335			   unsigned long base)
1336{
1337	struct tc_u_knode *n = fh;
1338
1339	tc_cls_bind_class(classid, cl, q, &n->res, base);
1340}
1341
1342static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1343		    struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1344{
1345	struct tc_u_knode *n = fh;
1346	struct tc_u_hnode *ht_up, *ht_down;
1347	struct nlattr *nest;
1348
1349	if (n == NULL)
1350		return skb->len;
1351
1352	t->tcm_handle = n->handle;
1353
1354	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1355	if (nest == NULL)
1356		goto nla_put_failure;
1357
1358	if (TC_U32_KEY(n->handle) == 0) {
1359		struct tc_u_hnode *ht = fh;
1360		u32 divisor = ht->divisor + 1;
1361
1362		if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1363			goto nla_put_failure;
1364	} else {
1365#ifdef CONFIG_CLS_U32_PERF
1366		struct tc_u32_pcnt *gpf;
1367		int cpu;
1368#endif
1369
1370		if (nla_put(skb, TCA_U32_SEL, struct_size(&n->sel, keys, n->sel.nkeys),
1371			    &n->sel))
1372			goto nla_put_failure;
1373
1374		ht_up = rtnl_dereference(n->ht_up);
1375		if (ht_up) {
1376			u32 htid = n->handle & 0xFFFFF000;
1377			if (nla_put_u32(skb, TCA_U32_HASH, htid))
1378				goto nla_put_failure;
1379		}
1380		if (n->res.classid &&
1381		    nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1382			goto nla_put_failure;
1383
1384		ht_down = rtnl_dereference(n->ht_down);
1385		if (ht_down &&
1386		    nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1387			goto nla_put_failure;
1388
1389		if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1390			goto nla_put_failure;
1391
1392#ifdef CONFIG_CLS_U32_MARK
1393		if ((n->val || n->mask)) {
1394			struct tc_u32_mark mark = {.val = n->val,
1395						   .mask = n->mask,
1396						   .success = 0};
1397			int cpum;
1398
1399			for_each_possible_cpu(cpum) {
1400				__u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1401
1402				mark.success += cnt;
1403			}
1404
1405			if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1406				goto nla_put_failure;
1407		}
1408#endif
1409
1410		if (tcf_exts_dump(skb, &n->exts) < 0)
1411			goto nla_put_failure;
1412
1413		if (n->ifindex) {
1414			struct net_device *dev;
1415			dev = __dev_get_by_index(net, n->ifindex);
1416			if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1417				goto nla_put_failure;
1418		}
1419#ifdef CONFIG_CLS_U32_PERF
1420		gpf = kzalloc(struct_size(gpf, kcnts, n->sel.nkeys), GFP_KERNEL);
1421		if (!gpf)
1422			goto nla_put_failure;
1423
1424		for_each_possible_cpu(cpu) {
1425			int i;
1426			struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1427
1428			gpf->rcnt += pf->rcnt;
1429			gpf->rhit += pf->rhit;
1430			for (i = 0; i < n->sel.nkeys; i++)
1431				gpf->kcnts[i] += pf->kcnts[i];
1432		}
1433
1434		if (nla_put_64bit(skb, TCA_U32_PCNT, struct_size(gpf, kcnts, n->sel.nkeys),
1435				  gpf, TCA_U32_PAD)) {
1436			kfree(gpf);
1437			goto nla_put_failure;
1438		}
1439		kfree(gpf);
1440#endif
1441	}
1442
1443	nla_nest_end(skb, nest);
1444
1445	if (TC_U32_KEY(n->handle))
1446		if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1447			goto nla_put_failure;
1448	return skb->len;
1449
1450nla_put_failure:
1451	nla_nest_cancel(skb, nest);
1452	return -1;
1453}
1454
1455static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1456	.kind		=	"u32",
1457	.classify	=	u32_classify,
1458	.init		=	u32_init,
1459	.destroy	=	u32_destroy,
1460	.get		=	u32_get,
1461	.change		=	u32_change,
1462	.delete		=	u32_delete,
1463	.walk		=	u32_walk,
1464	.reoffload	=	u32_reoffload,
1465	.dump		=	u32_dump,
1466	.bind_class	=	u32_bind_class,
1467	.owner		=	THIS_MODULE,
1468};
1469MODULE_ALIAS_NET_CLS("u32");
1470
1471static int __init init_u32(void)
1472{
1473	int i, ret;
1474
1475	pr_info("u32 classifier\n");
1476#ifdef CONFIG_CLS_U32_PERF
1477	pr_info("    Performance counters on\n");
1478#endif
1479	pr_info("    input device check on\n");
1480#ifdef CONFIG_NET_CLS_ACT
1481	pr_info("    Actions configured\n");
1482#endif
1483	tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1484					  sizeof(struct hlist_head),
1485					  GFP_KERNEL);
1486	if (!tc_u_common_hash)
1487		return -ENOMEM;
1488
1489	for (i = 0; i < U32_HASH_SIZE; i++)
1490		INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1491
1492	ret = register_tcf_proto_ops(&cls_u32_ops);
1493	if (ret)
1494		kvfree(tc_u_common_hash);
1495	return ret;
1496}
1497
1498static void __exit exit_u32(void)
1499{
1500	unregister_tcf_proto_ops(&cls_u32_ops);
1501	kvfree(tc_u_common_hash);
1502}
1503
1504module_init(init_u32)
1505module_exit(exit_u32)
1506MODULE_DESCRIPTION("Universal 32bit based TC Classifier");
1507MODULE_LICENSE("GPL");