xref: /openbmc/linux/net/sched/act_skbedit.c (revision 2bc7d3e0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2008, Intel Corporation.
4  *
5  * Author: Alexander Duyck <alexander.h.duyck@intel.com>
6  */
7 
8 #include <linux/module.h>
9 #include <linux/init.h>
10 #include <linux/kernel.h>
11 #include <linux/skbuff.h>
12 #include <linux/rtnetlink.h>
13 #include <net/netlink.h>
14 #include <net/pkt_sched.h>
15 #include <net/ip.h>
16 #include <net/ipv6.h>
17 #include <net/dsfield.h>
18 #include <net/pkt_cls.h>
19 
20 #include <linux/tc_act/tc_skbedit.h>
21 #include <net/tc_act/tc_skbedit.h>
22 
23 static unsigned int skbedit_net_id;
24 static struct tc_action_ops act_skbedit_ops;
25 
26 static u16 tcf_skbedit_hash(struct tcf_skbedit_params *params,
27 			    struct sk_buff *skb)
28 {
29 	u16 queue_mapping = params->queue_mapping;
30 
31 	if (params->flags & SKBEDIT_F_TXQ_SKBHASH) {
32 		u32 hash = skb_get_hash(skb);
33 
34 		queue_mapping += hash % params->mapping_mod;
35 	}
36 
37 	return netdev_cap_txqueue(skb->dev, queue_mapping);
38 }
39 
40 static int tcf_skbedit_act(struct sk_buff *skb, const struct tc_action *a,
41 			   struct tcf_result *res)
42 {
43 	struct tcf_skbedit *d = to_skbedit(a);
44 	struct tcf_skbedit_params *params;
45 	int action;
46 
47 	tcf_lastuse_update(&d->tcf_tm);
48 	bstats_update(this_cpu_ptr(d->common.cpu_bstats), skb);
49 
50 	params = rcu_dereference_bh(d->params);
51 	action = READ_ONCE(d->tcf_action);
52 
53 	if (params->flags & SKBEDIT_F_PRIORITY)
54 		skb->priority = params->priority;
55 	if (params->flags & SKBEDIT_F_INHERITDSFIELD) {
56 		int wlen = skb_network_offset(skb);
57 
58 		switch (skb_protocol(skb, true)) {
59 		case htons(ETH_P_IP):
60 			wlen += sizeof(struct iphdr);
61 			if (!pskb_may_pull(skb, wlen))
62 				goto err;
63 			skb->priority = ipv4_get_dsfield(ip_hdr(skb)) >> 2;
64 			break;
65 
66 		case htons(ETH_P_IPV6):
67 			wlen += sizeof(struct ipv6hdr);
68 			if (!pskb_may_pull(skb, wlen))
69 				goto err;
70 			skb->priority = ipv6_get_dsfield(ipv6_hdr(skb)) >> 2;
71 			break;
72 		}
73 	}
74 	if (params->flags & SKBEDIT_F_QUEUE_MAPPING &&
75 	    skb->dev->real_num_tx_queues > params->queue_mapping) {
76 #ifdef CONFIG_NET_EGRESS
77 		netdev_xmit_skip_txqueue(true);
78 #endif
79 		skb_set_queue_mapping(skb, tcf_skbedit_hash(params, skb));
80 	}
81 	if (params->flags & SKBEDIT_F_MARK) {
82 		skb->mark &= ~params->mask;
83 		skb->mark |= params->mark & params->mask;
84 	}
85 	if (params->flags & SKBEDIT_F_PTYPE)
86 		skb->pkt_type = params->ptype;
87 	return action;
88 
89 err:
90 	qstats_drop_inc(this_cpu_ptr(d->common.cpu_qstats));
91 	return TC_ACT_SHOT;
92 }
93 
94 static void tcf_skbedit_stats_update(struct tc_action *a, u64 bytes,
95 				     u64 packets, u64 drops,
96 				     u64 lastuse, bool hw)
97 {
98 	struct tcf_skbedit *d = to_skbedit(a);
99 	struct tcf_t *tm = &d->tcf_tm;
100 
101 	tcf_action_update_stats(a, bytes, packets, drops, hw);
102 	tm->lastuse = max_t(u64, tm->lastuse, lastuse);
103 }
104 
105 static const struct nla_policy skbedit_policy[TCA_SKBEDIT_MAX + 1] = {
106 	[TCA_SKBEDIT_PARMS]		= { .len = sizeof(struct tc_skbedit) },
107 	[TCA_SKBEDIT_PRIORITY]		= { .len = sizeof(u32) },
108 	[TCA_SKBEDIT_QUEUE_MAPPING]	= { .len = sizeof(u16) },
109 	[TCA_SKBEDIT_MARK]		= { .len = sizeof(u32) },
110 	[TCA_SKBEDIT_PTYPE]		= { .len = sizeof(u16) },
111 	[TCA_SKBEDIT_MASK]		= { .len = sizeof(u32) },
112 	[TCA_SKBEDIT_FLAGS]		= { .len = sizeof(u64) },
113 	[TCA_SKBEDIT_QUEUE_MAPPING_MAX]	= { .len = sizeof(u16) },
114 };
115 
116 static int tcf_skbedit_init(struct net *net, struct nlattr *nla,
117 			    struct nlattr *est, struct tc_action **a,
118 			    struct tcf_proto *tp, u32 act_flags,
119 			    struct netlink_ext_ack *extack)
120 {
121 	struct tc_action_net *tn = net_generic(net, skbedit_net_id);
122 	bool bind = act_flags & TCA_ACT_FLAGS_BIND;
123 	struct tcf_skbedit_params *params_new;
124 	struct nlattr *tb[TCA_SKBEDIT_MAX + 1];
125 	struct tcf_chain *goto_ch = NULL;
126 	struct tc_skbedit *parm;
127 	struct tcf_skbedit *d;
128 	u32 flags = 0, *priority = NULL, *mark = NULL, *mask = NULL;
129 	u16 *queue_mapping = NULL, *ptype = NULL;
130 	u16 mapping_mod = 1;
131 	bool exists = false;
132 	int ret = 0, err;
133 	u32 index;
134 
135 	if (nla == NULL)
136 		return -EINVAL;
137 
138 	err = nla_parse_nested_deprecated(tb, TCA_SKBEDIT_MAX, nla,
139 					  skbedit_policy, NULL);
140 	if (err < 0)
141 		return err;
142 
143 	if (tb[TCA_SKBEDIT_PARMS] == NULL)
144 		return -EINVAL;
145 
146 	if (tb[TCA_SKBEDIT_PRIORITY] != NULL) {
147 		flags |= SKBEDIT_F_PRIORITY;
148 		priority = nla_data(tb[TCA_SKBEDIT_PRIORITY]);
149 	}
150 
151 	if (tb[TCA_SKBEDIT_QUEUE_MAPPING] != NULL) {
152 		flags |= SKBEDIT_F_QUEUE_MAPPING;
153 		queue_mapping = nla_data(tb[TCA_SKBEDIT_QUEUE_MAPPING]);
154 	}
155 
156 	if (tb[TCA_SKBEDIT_PTYPE] != NULL) {
157 		ptype = nla_data(tb[TCA_SKBEDIT_PTYPE]);
158 		if (!skb_pkt_type_ok(*ptype))
159 			return -EINVAL;
160 		flags |= SKBEDIT_F_PTYPE;
161 	}
162 
163 	if (tb[TCA_SKBEDIT_MARK] != NULL) {
164 		flags |= SKBEDIT_F_MARK;
165 		mark = nla_data(tb[TCA_SKBEDIT_MARK]);
166 	}
167 
168 	if (tb[TCA_SKBEDIT_MASK] != NULL) {
169 		flags |= SKBEDIT_F_MASK;
170 		mask = nla_data(tb[TCA_SKBEDIT_MASK]);
171 	}
172 
173 	if (tb[TCA_SKBEDIT_FLAGS] != NULL) {
174 		u64 *pure_flags = nla_data(tb[TCA_SKBEDIT_FLAGS]);
175 
176 		if (*pure_flags & SKBEDIT_F_TXQ_SKBHASH) {
177 			u16 *queue_mapping_max;
178 
179 			if (!tb[TCA_SKBEDIT_QUEUE_MAPPING] ||
180 			    !tb[TCA_SKBEDIT_QUEUE_MAPPING_MAX]) {
181 				NL_SET_ERR_MSG_MOD(extack, "Missing required range of queue_mapping.");
182 				return -EINVAL;
183 			}
184 
185 			queue_mapping_max =
186 				nla_data(tb[TCA_SKBEDIT_QUEUE_MAPPING_MAX]);
187 			if (*queue_mapping_max < *queue_mapping) {
188 				NL_SET_ERR_MSG_MOD(extack, "The range of queue_mapping is invalid, max < min.");
189 				return -EINVAL;
190 			}
191 
192 			mapping_mod = *queue_mapping_max - *queue_mapping + 1;
193 			flags |= SKBEDIT_F_TXQ_SKBHASH;
194 		}
195 		if (*pure_flags & SKBEDIT_F_INHERITDSFIELD)
196 			flags |= SKBEDIT_F_INHERITDSFIELD;
197 	}
198 
199 	parm = nla_data(tb[TCA_SKBEDIT_PARMS]);
200 	index = parm->index;
201 	err = tcf_idr_check_alloc(tn, &index, a, bind);
202 	if (err < 0)
203 		return err;
204 	exists = err;
205 	if (exists && bind)
206 		return 0;
207 
208 	if (!flags) {
209 		if (exists)
210 			tcf_idr_release(*a, bind);
211 		else
212 			tcf_idr_cleanup(tn, index);
213 		return -EINVAL;
214 	}
215 
216 	if (!exists) {
217 		ret = tcf_idr_create(tn, index, est, a,
218 				     &act_skbedit_ops, bind, true, act_flags);
219 		if (ret) {
220 			tcf_idr_cleanup(tn, index);
221 			return ret;
222 		}
223 
224 		d = to_skbedit(*a);
225 		ret = ACT_P_CREATED;
226 	} else {
227 		d = to_skbedit(*a);
228 		if (!(act_flags & TCA_ACT_FLAGS_REPLACE)) {
229 			tcf_idr_release(*a, bind);
230 			return -EEXIST;
231 		}
232 	}
233 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
234 	if (err < 0)
235 		goto release_idr;
236 
237 	params_new = kzalloc(sizeof(*params_new), GFP_KERNEL);
238 	if (unlikely(!params_new)) {
239 		err = -ENOMEM;
240 		goto put_chain;
241 	}
242 
243 	params_new->flags = flags;
244 	if (flags & SKBEDIT_F_PRIORITY)
245 		params_new->priority = *priority;
246 	if (flags & SKBEDIT_F_QUEUE_MAPPING) {
247 		params_new->queue_mapping = *queue_mapping;
248 		params_new->mapping_mod = mapping_mod;
249 	}
250 	if (flags & SKBEDIT_F_MARK)
251 		params_new->mark = *mark;
252 	if (flags & SKBEDIT_F_PTYPE)
253 		params_new->ptype = *ptype;
254 	/* default behaviour is to use all the bits */
255 	params_new->mask = 0xffffffff;
256 	if (flags & SKBEDIT_F_MASK)
257 		params_new->mask = *mask;
258 
259 	spin_lock_bh(&d->tcf_lock);
260 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
261 	params_new = rcu_replace_pointer(d->params, params_new,
262 					 lockdep_is_held(&d->tcf_lock));
263 	spin_unlock_bh(&d->tcf_lock);
264 	if (params_new)
265 		kfree_rcu(params_new, rcu);
266 	if (goto_ch)
267 		tcf_chain_put_by_act(goto_ch);
268 
269 	return ret;
270 put_chain:
271 	if (goto_ch)
272 		tcf_chain_put_by_act(goto_ch);
273 release_idr:
274 	tcf_idr_release(*a, bind);
275 	return err;
276 }
277 
278 static int tcf_skbedit_dump(struct sk_buff *skb, struct tc_action *a,
279 			    int bind, int ref)
280 {
281 	unsigned char *b = skb_tail_pointer(skb);
282 	struct tcf_skbedit *d = to_skbedit(a);
283 	struct tcf_skbedit_params *params;
284 	struct tc_skbedit opt = {
285 		.index   = d->tcf_index,
286 		.refcnt  = refcount_read(&d->tcf_refcnt) - ref,
287 		.bindcnt = atomic_read(&d->tcf_bindcnt) - bind,
288 	};
289 	u64 pure_flags = 0;
290 	struct tcf_t t;
291 
292 	spin_lock_bh(&d->tcf_lock);
293 	params = rcu_dereference_protected(d->params,
294 					   lockdep_is_held(&d->tcf_lock));
295 	opt.action = d->tcf_action;
296 
297 	if (nla_put(skb, TCA_SKBEDIT_PARMS, sizeof(opt), &opt))
298 		goto nla_put_failure;
299 	if ((params->flags & SKBEDIT_F_PRIORITY) &&
300 	    nla_put_u32(skb, TCA_SKBEDIT_PRIORITY, params->priority))
301 		goto nla_put_failure;
302 	if ((params->flags & SKBEDIT_F_QUEUE_MAPPING) &&
303 	    nla_put_u16(skb, TCA_SKBEDIT_QUEUE_MAPPING, params->queue_mapping))
304 		goto nla_put_failure;
305 	if ((params->flags & SKBEDIT_F_MARK) &&
306 	    nla_put_u32(skb, TCA_SKBEDIT_MARK, params->mark))
307 		goto nla_put_failure;
308 	if ((params->flags & SKBEDIT_F_PTYPE) &&
309 	    nla_put_u16(skb, TCA_SKBEDIT_PTYPE, params->ptype))
310 		goto nla_put_failure;
311 	if ((params->flags & SKBEDIT_F_MASK) &&
312 	    nla_put_u32(skb, TCA_SKBEDIT_MASK, params->mask))
313 		goto nla_put_failure;
314 	if (params->flags & SKBEDIT_F_INHERITDSFIELD)
315 		pure_flags |= SKBEDIT_F_INHERITDSFIELD;
316 	if (params->flags & SKBEDIT_F_TXQ_SKBHASH) {
317 		if (nla_put_u16(skb, TCA_SKBEDIT_QUEUE_MAPPING_MAX,
318 				params->queue_mapping + params->mapping_mod - 1))
319 			goto nla_put_failure;
320 
321 		pure_flags |= SKBEDIT_F_TXQ_SKBHASH;
322 	}
323 	if (pure_flags != 0 &&
324 	    nla_put(skb, TCA_SKBEDIT_FLAGS, sizeof(pure_flags), &pure_flags))
325 		goto nla_put_failure;
326 
327 	tcf_tm_dump(&t, &d->tcf_tm);
328 	if (nla_put_64bit(skb, TCA_SKBEDIT_TM, sizeof(t), &t, TCA_SKBEDIT_PAD))
329 		goto nla_put_failure;
330 	spin_unlock_bh(&d->tcf_lock);
331 
332 	return skb->len;
333 
334 nla_put_failure:
335 	spin_unlock_bh(&d->tcf_lock);
336 	nlmsg_trim(skb, b);
337 	return -1;
338 }
339 
340 static void tcf_skbedit_cleanup(struct tc_action *a)
341 {
342 	struct tcf_skbedit *d = to_skbedit(a);
343 	struct tcf_skbedit_params *params;
344 
345 	params = rcu_dereference_protected(d->params, 1);
346 	if (params)
347 		kfree_rcu(params, rcu);
348 }
349 
350 static int tcf_skbedit_walker(struct net *net, struct sk_buff *skb,
351 			      struct netlink_callback *cb, int type,
352 			      const struct tc_action_ops *ops,
353 			      struct netlink_ext_ack *extack)
354 {
355 	struct tc_action_net *tn = net_generic(net, skbedit_net_id);
356 
357 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
358 }
359 
360 static int tcf_skbedit_search(struct net *net, struct tc_action **a, u32 index)
361 {
362 	struct tc_action_net *tn = net_generic(net, skbedit_net_id);
363 
364 	return tcf_idr_search(tn, a, index);
365 }
366 
367 static size_t tcf_skbedit_get_fill_size(const struct tc_action *act)
368 {
369 	return nla_total_size(sizeof(struct tc_skbedit))
370 		+ nla_total_size(sizeof(u32)) /* TCA_SKBEDIT_PRIORITY */
371 		+ nla_total_size(sizeof(u16)) /* TCA_SKBEDIT_QUEUE_MAPPING */
372 		+ nla_total_size(sizeof(u16)) /* TCA_SKBEDIT_QUEUE_MAPPING_MAX */
373 		+ nla_total_size(sizeof(u32)) /* TCA_SKBEDIT_MARK */
374 		+ nla_total_size(sizeof(u16)) /* TCA_SKBEDIT_PTYPE */
375 		+ nla_total_size(sizeof(u32)) /* TCA_SKBEDIT_MASK */
376 		+ nla_total_size_64bit(sizeof(u64)); /* TCA_SKBEDIT_FLAGS */
377 }
378 
379 static int tcf_skbedit_offload_act_setup(struct tc_action *act, void *entry_data,
380 					 u32 *index_inc, bool bind,
381 					 struct netlink_ext_ack *extack)
382 {
383 	if (bind) {
384 		struct flow_action_entry *entry = entry_data;
385 
386 		if (is_tcf_skbedit_mark(act)) {
387 			entry->id = FLOW_ACTION_MARK;
388 			entry->mark = tcf_skbedit_mark(act);
389 		} else if (is_tcf_skbedit_ptype(act)) {
390 			entry->id = FLOW_ACTION_PTYPE;
391 			entry->ptype = tcf_skbedit_ptype(act);
392 		} else if (is_tcf_skbedit_priority(act)) {
393 			entry->id = FLOW_ACTION_PRIORITY;
394 			entry->priority = tcf_skbedit_priority(act);
395 		} else if (is_tcf_skbedit_queue_mapping(act)) {
396 			NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"queue_mapping\" option is used");
397 			return -EOPNOTSUPP;
398 		} else if (is_tcf_skbedit_inheritdsfield(act)) {
399 			NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"inheritdsfield\" option is used");
400 			return -EOPNOTSUPP;
401 		} else {
402 			NL_SET_ERR_MSG_MOD(extack, "Unsupported skbedit option offload");
403 			return -EOPNOTSUPP;
404 		}
405 		*index_inc = 1;
406 	} else {
407 		struct flow_offload_action *fl_action = entry_data;
408 
409 		if (is_tcf_skbedit_mark(act))
410 			fl_action->id = FLOW_ACTION_MARK;
411 		else if (is_tcf_skbedit_ptype(act))
412 			fl_action->id = FLOW_ACTION_PTYPE;
413 		else if (is_tcf_skbedit_priority(act))
414 			fl_action->id = FLOW_ACTION_PRIORITY;
415 		else
416 			return -EOPNOTSUPP;
417 	}
418 
419 	return 0;
420 }
421 
422 static struct tc_action_ops act_skbedit_ops = {
423 	.kind		=	"skbedit",
424 	.id		=	TCA_ID_SKBEDIT,
425 	.owner		=	THIS_MODULE,
426 	.act		=	tcf_skbedit_act,
427 	.stats_update	=	tcf_skbedit_stats_update,
428 	.dump		=	tcf_skbedit_dump,
429 	.init		=	tcf_skbedit_init,
430 	.cleanup	=	tcf_skbedit_cleanup,
431 	.walk		=	tcf_skbedit_walker,
432 	.get_fill_size	=	tcf_skbedit_get_fill_size,
433 	.lookup		=	tcf_skbedit_search,
434 	.offload_act_setup =	tcf_skbedit_offload_act_setup,
435 	.size		=	sizeof(struct tcf_skbedit),
436 };
437 
438 static __net_init int skbedit_init_net(struct net *net)
439 {
440 	struct tc_action_net *tn = net_generic(net, skbedit_net_id);
441 
442 	return tc_action_net_init(net, tn, &act_skbedit_ops);
443 }
444 
445 static void __net_exit skbedit_exit_net(struct list_head *net_list)
446 {
447 	tc_action_net_exit(net_list, skbedit_net_id);
448 }
449 
450 static struct pernet_operations skbedit_net_ops = {
451 	.init = skbedit_init_net,
452 	.exit_batch = skbedit_exit_net,
453 	.id   = &skbedit_net_id,
454 	.size = sizeof(struct tc_action_net),
455 };
456 
457 MODULE_AUTHOR("Alexander Duyck, <alexander.h.duyck@intel.com>");
458 MODULE_DESCRIPTION("SKB Editing");
459 MODULE_LICENSE("GPL");
460 
461 static int __init skbedit_init_module(void)
462 {
463 	return tcf_register_action(&act_skbedit_ops, &skbedit_net_ops);
464 }
465 
466 static void __exit skbedit_cleanup_module(void)
467 {
468 	tcf_unregister_action(&act_skbedit_ops, &skbedit_net_ops);
469 }
470 
471 module_init(skbedit_init_module);
472 module_exit(skbedit_cleanup_module);
473