xref: /linux/net/netfilter/nf_tables_core.c (revision d6fd48ef)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2008 Patrick McHardy <kaber@trash.net>
4  *
5  * Development of this code funded by Astaro AG (http://www.astaro.com/)
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/list.h>
12 #include <linux/rculist.h>
13 #include <linux/skbuff.h>
14 #include <linux/netlink.h>
15 #include <linux/netfilter.h>
16 #include <linux/static_key.h>
17 #include <linux/netfilter/nfnetlink.h>
18 #include <linux/netfilter/nf_tables.h>
19 #include <net/netfilter/nf_tables_core.h>
20 #include <net/netfilter/nf_tables.h>
21 #include <net/netfilter/nf_log.h>
22 #include <net/netfilter/nft_meta.h>
23 
24 #if defined(CONFIG_RETPOLINE) && defined(CONFIG_X86)
25 
26 static struct static_key_false nf_tables_skip_direct_calls;
27 
28 static bool nf_skip_indirect_calls(void)
29 {
30 	return static_branch_likely(&nf_tables_skip_direct_calls);
31 }
32 
33 static void __init nf_skip_indirect_calls_enable(void)
34 {
35 	if (!cpu_feature_enabled(X86_FEATURE_RETPOLINE))
36 		static_branch_enable(&nf_tables_skip_direct_calls);
37 }
38 #else
39 static inline bool nf_skip_indirect_calls(void) { return false; }
40 
41 static inline void nf_skip_indirect_calls_enable(void) { }
42 #endif
43 
44 static noinline void __nft_trace_packet(struct nft_traceinfo *info,
45 					const struct nft_chain *chain,
46 					enum nft_trace_types type)
47 {
48 	if (!info->trace || !info->nf_trace)
49 		return;
50 
51 	info->chain = chain;
52 	info->type = type;
53 
54 	nft_trace_notify(info);
55 }
56 
57 static inline void nft_trace_packet(const struct nft_pktinfo *pkt,
58 				    struct nft_traceinfo *info,
59 				    const struct nft_chain *chain,
60 				    const struct nft_rule_dp *rule,
61 				    enum nft_trace_types type)
62 {
63 	if (static_branch_unlikely(&nft_trace_enabled)) {
64 		info->nf_trace = pkt->skb->nf_trace;
65 		info->rule = rule;
66 		__nft_trace_packet(info, chain, type);
67 	}
68 }
69 
70 static inline void nft_trace_copy_nftrace(const struct nft_pktinfo *pkt,
71 					  struct nft_traceinfo *info)
72 {
73 	if (static_branch_unlikely(&nft_trace_enabled)) {
74 		if (info->trace)
75 			info->nf_trace = pkt->skb->nf_trace;
76 	}
77 }
78 
79 static void nft_bitwise_fast_eval(const struct nft_expr *expr,
80 				  struct nft_regs *regs)
81 {
82 	const struct nft_bitwise_fast_expr *priv = nft_expr_priv(expr);
83 	u32 *src = &regs->data[priv->sreg];
84 	u32 *dst = &regs->data[priv->dreg];
85 
86 	*dst = (*src & priv->mask) ^ priv->xor;
87 }
88 
89 static void nft_cmp_fast_eval(const struct nft_expr *expr,
90 			      struct nft_regs *regs)
91 {
92 	const struct nft_cmp_fast_expr *priv = nft_expr_priv(expr);
93 
94 	if (((regs->data[priv->sreg] & priv->mask) == priv->data) ^ priv->inv)
95 		return;
96 	regs->verdict.code = NFT_BREAK;
97 }
98 
99 static void nft_cmp16_fast_eval(const struct nft_expr *expr,
100 				struct nft_regs *regs)
101 {
102 	const struct nft_cmp16_fast_expr *priv = nft_expr_priv(expr);
103 	const u64 *reg_data = (const u64 *)&regs->data[priv->sreg];
104 	const u64 *mask = (const u64 *)&priv->mask;
105 	const u64 *data = (const u64 *)&priv->data;
106 
107 	if (((reg_data[0] & mask[0]) == data[0] &&
108 	    ((reg_data[1] & mask[1]) == data[1])) ^ priv->inv)
109 		return;
110 	regs->verdict.code = NFT_BREAK;
111 }
112 
113 static noinline void __nft_trace_verdict(struct nft_traceinfo *info,
114 					 const struct nft_chain *chain,
115 					 const struct nft_regs *regs)
116 {
117 	enum nft_trace_types type;
118 
119 	switch (regs->verdict.code) {
120 	case NFT_CONTINUE:
121 	case NFT_RETURN:
122 		type = NFT_TRACETYPE_RETURN;
123 		break;
124 	case NF_STOLEN:
125 		type = NFT_TRACETYPE_RULE;
126 		/* can't access skb->nf_trace; use copy */
127 		break;
128 	default:
129 		type = NFT_TRACETYPE_RULE;
130 
131 		if (info->trace)
132 			info->nf_trace = info->pkt->skb->nf_trace;
133 		break;
134 	}
135 
136 	__nft_trace_packet(info, chain, type);
137 }
138 
139 static inline void nft_trace_verdict(struct nft_traceinfo *info,
140 				     const struct nft_chain *chain,
141 				     const struct nft_rule_dp *rule,
142 				     const struct nft_regs *regs)
143 {
144 	if (static_branch_unlikely(&nft_trace_enabled)) {
145 		info->rule = rule;
146 		__nft_trace_verdict(info, chain, regs);
147 	}
148 }
149 
150 static bool nft_payload_fast_eval(const struct nft_expr *expr,
151 				  struct nft_regs *regs,
152 				  const struct nft_pktinfo *pkt)
153 {
154 	const struct nft_payload *priv = nft_expr_priv(expr);
155 	const struct sk_buff *skb = pkt->skb;
156 	u32 *dest = &regs->data[priv->dreg];
157 	unsigned char *ptr;
158 
159 	if (priv->base == NFT_PAYLOAD_NETWORK_HEADER)
160 		ptr = skb_network_header(skb);
161 	else {
162 		if (!(pkt->flags & NFT_PKTINFO_L4PROTO))
163 			return false;
164 		ptr = skb_network_header(skb) + nft_thoff(pkt);
165 	}
166 
167 	ptr += priv->offset;
168 
169 	if (unlikely(ptr + priv->len > skb_tail_pointer(skb)))
170 		return false;
171 
172 	*dest = 0;
173 	if (priv->len == 2)
174 		*(u16 *)dest = *(u16 *)ptr;
175 	else if (priv->len == 4)
176 		*(u32 *)dest = *(u32 *)ptr;
177 	else
178 		*(u8 *)dest = *(u8 *)ptr;
179 	return true;
180 }
181 
182 DEFINE_STATIC_KEY_FALSE(nft_counters_enabled);
183 
184 static noinline void nft_update_chain_stats(const struct nft_chain *chain,
185 					    const struct nft_pktinfo *pkt)
186 {
187 	struct nft_base_chain *base_chain;
188 	struct nft_stats __percpu *pstats;
189 	struct nft_stats *stats;
190 
191 	base_chain = nft_base_chain(chain);
192 
193 	pstats = READ_ONCE(base_chain->stats);
194 	if (pstats) {
195 		local_bh_disable();
196 		stats = this_cpu_ptr(pstats);
197 		u64_stats_update_begin(&stats->syncp);
198 		stats->pkts++;
199 		stats->bytes += pkt->skb->len;
200 		u64_stats_update_end(&stats->syncp);
201 		local_bh_enable();
202 	}
203 }
204 
205 struct nft_jumpstack {
206 	const struct nft_chain *chain;
207 	const struct nft_rule_dp *rule;
208 	const struct nft_rule_dp *last_rule;
209 };
210 
211 static void expr_call_ops_eval(const struct nft_expr *expr,
212 			       struct nft_regs *regs,
213 			       struct nft_pktinfo *pkt)
214 {
215 #ifdef CONFIG_RETPOLINE
216 	unsigned long e;
217 
218 	if (nf_skip_indirect_calls())
219 		goto indirect_call;
220 
221 	e = (unsigned long)expr->ops->eval;
222 #define X(e, fun) \
223 	do { if ((e) == (unsigned long)(fun)) \
224 		return fun(expr, regs, pkt); } while (0)
225 
226 	X(e, nft_payload_eval);
227 	X(e, nft_cmp_eval);
228 	X(e, nft_counter_eval);
229 	X(e, nft_meta_get_eval);
230 	X(e, nft_lookup_eval);
231 #if IS_ENABLED(CONFIG_NFT_CT)
232 	X(e, nft_ct_get_fast_eval);
233 #endif
234 	X(e, nft_range_eval);
235 	X(e, nft_immediate_eval);
236 	X(e, nft_byteorder_eval);
237 	X(e, nft_dynset_eval);
238 	X(e, nft_rt_get_eval);
239 	X(e, nft_bitwise_eval);
240 	X(e, nft_objref_eval);
241 	X(e, nft_objref_map_eval);
242 #undef  X
243 indirect_call:
244 #endif /* CONFIG_RETPOLINE */
245 	expr->ops->eval(expr, regs, pkt);
246 }
247 
248 #define nft_rule_expr_first(rule)	(struct nft_expr *)&rule->data[0]
249 #define nft_rule_expr_next(expr)	((void *)expr) + expr->ops->size
250 #define nft_rule_expr_last(rule)	(struct nft_expr *)&rule->data[rule->dlen]
251 #define nft_rule_next(rule)		(void *)rule + sizeof(*rule) + rule->dlen
252 
253 #define nft_rule_dp_for_each_expr(expr, last, rule) \
254         for ((expr) = nft_rule_expr_first(rule), (last) = nft_rule_expr_last(rule); \
255              (expr) != (last); \
256              (expr) = nft_rule_expr_next(expr))
257 
258 unsigned int
259 nft_do_chain(struct nft_pktinfo *pkt, void *priv)
260 {
261 	const struct nft_chain *chain = priv, *basechain = chain;
262 	const struct nft_rule_dp *rule, *last_rule;
263 	const struct net *net = nft_net(pkt);
264 	const struct nft_expr *expr, *last;
265 	struct nft_regs regs = {};
266 	unsigned int stackptr = 0;
267 	struct nft_jumpstack jumpstack[NFT_JUMP_STACK_SIZE];
268 	bool genbit = READ_ONCE(net->nft.gencursor);
269 	struct nft_rule_blob *blob;
270 	struct nft_traceinfo info;
271 
272 	info.trace = false;
273 	if (static_branch_unlikely(&nft_trace_enabled))
274 		nft_trace_init(&info, pkt, &regs.verdict, basechain);
275 do_chain:
276 	if (genbit)
277 		blob = rcu_dereference(chain->blob_gen_1);
278 	else
279 		blob = rcu_dereference(chain->blob_gen_0);
280 
281 	rule = (struct nft_rule_dp *)blob->data;
282 	last_rule = (void *)blob->data + blob->size;
283 next_rule:
284 	regs.verdict.code = NFT_CONTINUE;
285 	for (; rule < last_rule; rule = nft_rule_next(rule)) {
286 		nft_rule_dp_for_each_expr(expr, last, rule) {
287 			if (expr->ops == &nft_cmp_fast_ops)
288 				nft_cmp_fast_eval(expr, &regs);
289 			else if (expr->ops == &nft_cmp16_fast_ops)
290 				nft_cmp16_fast_eval(expr, &regs);
291 			else if (expr->ops == &nft_bitwise_fast_ops)
292 				nft_bitwise_fast_eval(expr, &regs);
293 			else if (expr->ops != &nft_payload_fast_ops ||
294 				 !nft_payload_fast_eval(expr, &regs, pkt))
295 				expr_call_ops_eval(expr, &regs, pkt);
296 
297 			if (regs.verdict.code != NFT_CONTINUE)
298 				break;
299 		}
300 
301 		switch (regs.verdict.code) {
302 		case NFT_BREAK:
303 			regs.verdict.code = NFT_CONTINUE;
304 			nft_trace_copy_nftrace(pkt, &info);
305 			continue;
306 		case NFT_CONTINUE:
307 			nft_trace_packet(pkt, &info, chain, rule,
308 					 NFT_TRACETYPE_RULE);
309 			continue;
310 		}
311 		break;
312 	}
313 
314 	nft_trace_verdict(&info, chain, rule, &regs);
315 
316 	switch (regs.verdict.code & NF_VERDICT_MASK) {
317 	case NF_ACCEPT:
318 	case NF_DROP:
319 	case NF_QUEUE:
320 	case NF_STOLEN:
321 		return regs.verdict.code;
322 	}
323 
324 	switch (regs.verdict.code) {
325 	case NFT_JUMP:
326 		if (WARN_ON_ONCE(stackptr >= NFT_JUMP_STACK_SIZE))
327 			return NF_DROP;
328 		jumpstack[stackptr].chain = chain;
329 		jumpstack[stackptr].rule = nft_rule_next(rule);
330 		jumpstack[stackptr].last_rule = last_rule;
331 		stackptr++;
332 		fallthrough;
333 	case NFT_GOTO:
334 		chain = regs.verdict.chain;
335 		goto do_chain;
336 	case NFT_CONTINUE:
337 	case NFT_RETURN:
338 		break;
339 	default:
340 		WARN_ON_ONCE(1);
341 	}
342 
343 	if (stackptr > 0) {
344 		stackptr--;
345 		chain = jumpstack[stackptr].chain;
346 		rule = jumpstack[stackptr].rule;
347 		last_rule = jumpstack[stackptr].last_rule;
348 		goto next_rule;
349 	}
350 
351 	nft_trace_packet(pkt, &info, basechain, NULL, NFT_TRACETYPE_POLICY);
352 
353 	if (static_branch_unlikely(&nft_counters_enabled))
354 		nft_update_chain_stats(basechain, pkt);
355 
356 	return nft_base_chain(basechain)->policy;
357 }
358 EXPORT_SYMBOL_GPL(nft_do_chain);
359 
360 static struct nft_expr_type *nft_basic_types[] = {
361 	&nft_imm_type,
362 	&nft_cmp_type,
363 	&nft_lookup_type,
364 	&nft_bitwise_type,
365 	&nft_byteorder_type,
366 	&nft_payload_type,
367 	&nft_dynset_type,
368 	&nft_range_type,
369 	&nft_meta_type,
370 	&nft_rt_type,
371 	&nft_exthdr_type,
372 	&nft_last_type,
373 	&nft_counter_type,
374 	&nft_objref_type,
375 	&nft_inner_type,
376 };
377 
378 static struct nft_object_type *nft_basic_objects[] = {
379 #ifdef CONFIG_NETWORK_SECMARK
380 	&nft_secmark_obj_type,
381 #endif
382 	&nft_counter_obj_type,
383 };
384 
385 int __init nf_tables_core_module_init(void)
386 {
387 	int err, i, j = 0;
388 
389 	nft_counter_init_seqcount();
390 
391 	for (i = 0; i < ARRAY_SIZE(nft_basic_objects); i++) {
392 		err = nft_register_obj(nft_basic_objects[i]);
393 		if (err)
394 			goto err;
395 	}
396 
397 	for (j = 0; j < ARRAY_SIZE(nft_basic_types); j++) {
398 		err = nft_register_expr(nft_basic_types[j]);
399 		if (err)
400 			goto err;
401 	}
402 
403 	nf_skip_indirect_calls_enable();
404 
405 	return 0;
406 
407 err:
408 	while (j-- > 0)
409 		nft_unregister_expr(nft_basic_types[j]);
410 
411 	while (i-- > 0)
412 		nft_unregister_obj(nft_basic_objects[i]);
413 
414 	return err;
415 }
416 
417 void nf_tables_core_module_exit(void)
418 {
419 	int i;
420 
421 	i = ARRAY_SIZE(nft_basic_types);
422 	while (i-- > 0)
423 		nft_unregister_expr(nft_basic_types[i]);
424 
425 	i = ARRAY_SIZE(nft_basic_objects);
426 	while (i-- > 0)
427 		nft_unregister_obj(nft_basic_objects[i]);
428 }
429