xref: /openbmc/linux/net/sched/act_skbedit.c (revision 1edd0337)
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 #include <net/tc_wrapper.h>
20 
21 #include <linux/tc_act/tc_skbedit.h>
22 #include <net/tc_act/tc_skbedit.h>
23 
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 TC_INDIRECT_SCOPE int tcf_skbedit_act(struct sk_buff *skb,
41 				      const struct tc_action *a,
42 				      struct tcf_result *res)
43 {
44 	struct tcf_skbedit *d = to_skbedit(a);
45 	struct tcf_skbedit_params *params;
46 	int action;
47 
48 	tcf_lastuse_update(&d->tcf_tm);
49 	bstats_update(this_cpu_ptr(d->common.cpu_bstats), skb);
50 
51 	params = rcu_dereference_bh(d->params);
52 	action = READ_ONCE(d->tcf_action);
53 
54 	if (params->flags & SKBEDIT_F_PRIORITY)
55 		skb->priority = params->priority;
56 	if (params->flags & SKBEDIT_F_INHERITDSFIELD) {
57 		int wlen = skb_network_offset(skb);
58 
59 		switch (skb_protocol(skb, true)) {
60 		case htons(ETH_P_IP):
61 			wlen += sizeof(struct iphdr);
62 			if (!pskb_may_pull(skb, wlen))
63 				goto err;
64 			skb->priority = ipv4_get_dsfield(ip_hdr(skb)) >> 2;
65 			break;
66 
67 		case htons(ETH_P_IPV6):
68 			wlen += sizeof(struct ipv6hdr);
69 			if (!pskb_may_pull(skb, wlen))
70 				goto err;
71 			skb->priority = ipv6_get_dsfield(ipv6_hdr(skb)) >> 2;
72 			break;
73 		}
74 	}
75 	if (params->flags & SKBEDIT_F_QUEUE_MAPPING &&
76 	    skb->dev->real_num_tx_queues > params->queue_mapping) {
77 #ifdef CONFIG_NET_EGRESS
78 		netdev_xmit_skip_txqueue(true);
79 #endif
80 		skb_set_queue_mapping(skb, tcf_skbedit_hash(params, skb));
81 	}
82 	if (params->flags & SKBEDIT_F_MARK) {
83 		skb->mark &= ~params->mask;
84 		skb->mark |= params->mark & params->mask;
85 	}
86 	if (params->flags & SKBEDIT_F_PTYPE)
87 		skb->pkt_type = params->ptype;
88 	return action;
89 
90 err:
91 	qstats_drop_inc(this_cpu_ptr(d->common.cpu_qstats));
92 	return TC_ACT_SHOT;
93 }
94 
95 static void tcf_skbedit_stats_update(struct tc_action *a, u64 bytes,
96 				     u64 packets, u64 drops,
97 				     u64 lastuse, bool hw)
98 {
99 	struct tcf_skbedit *d = to_skbedit(a);
100 	struct tcf_t *tm = &d->tcf_tm;
101 
102 	tcf_action_update_stats(a, bytes, packets, drops, hw);
103 	tm->lastuse = max_t(u64, tm->lastuse, lastuse);
104 }
105 
106 static const struct nla_policy skbedit_policy[TCA_SKBEDIT_MAX + 1] = {
107 	[TCA_SKBEDIT_PARMS]		= { .len = sizeof(struct tc_skbedit) },
108 	[TCA_SKBEDIT_PRIORITY]		= { .len = sizeof(u32) },
109 	[TCA_SKBEDIT_QUEUE_MAPPING]	= { .len = sizeof(u16) },
110 	[TCA_SKBEDIT_MARK]		= { .len = sizeof(u32) },
111 	[TCA_SKBEDIT_PTYPE]		= { .len = sizeof(u16) },
112 	[TCA_SKBEDIT_MASK]		= { .len = sizeof(u32) },
113 	[TCA_SKBEDIT_FLAGS]		= { .len = sizeof(u64) },
114 	[TCA_SKBEDIT_QUEUE_MAPPING_MAX]	= { .len = sizeof(u16) },
115 };
116 
117 static int tcf_skbedit_init(struct net *net, struct nlattr *nla,
118 			    struct nlattr *est, struct tc_action **a,
119 			    struct tcf_proto *tp, u32 act_flags,
120 			    struct netlink_ext_ack *extack)
121 {
122 	struct tc_action_net *tn = net_generic(net, act_skbedit_ops.net_id);
123 	bool bind = act_flags & TCA_ACT_FLAGS_BIND;
124 	struct tcf_skbedit_params *params_new;
125 	struct nlattr *tb[TCA_SKBEDIT_MAX + 1];
126 	struct tcf_chain *goto_ch = NULL;
127 	struct tc_skbedit *parm;
128 	struct tcf_skbedit *d;
129 	u32 flags = 0, *priority = NULL, *mark = NULL, *mask = NULL;
130 	u16 *queue_mapping = NULL, *ptype = NULL;
131 	u16 mapping_mod = 1;
132 	bool exists = false;
133 	int ret = 0, err;
134 	u32 index;
135 
136 	if (nla == NULL)
137 		return -EINVAL;
138 
139 	err = nla_parse_nested_deprecated(tb, TCA_SKBEDIT_MAX, nla,
140 					  skbedit_policy, NULL);
141 	if (err < 0)
142 		return err;
143 
144 	if (tb[TCA_SKBEDIT_PARMS] == NULL)
145 		return -EINVAL;
146 
147 	if (tb[TCA_SKBEDIT_PRIORITY] != NULL) {
148 		flags |= SKBEDIT_F_PRIORITY;
149 		priority = nla_data(tb[TCA_SKBEDIT_PRIORITY]);
150 	}
151 
152 	if (tb[TCA_SKBEDIT_QUEUE_MAPPING] != NULL) {
153 		if (is_tcf_skbedit_ingress(act_flags) &&
154 		    !(act_flags & TCA_ACT_FLAGS_SKIP_SW)) {
155 			NL_SET_ERR_MSG_MOD(extack, "\"queue_mapping\" option on receive side is hardware only, use skip_sw");
156 			return -EOPNOTSUPP;
157 		}
158 		flags |= SKBEDIT_F_QUEUE_MAPPING;
159 		queue_mapping = nla_data(tb[TCA_SKBEDIT_QUEUE_MAPPING]);
160 	}
161 
162 	if (tb[TCA_SKBEDIT_PTYPE] != NULL) {
163 		ptype = nla_data(tb[TCA_SKBEDIT_PTYPE]);
164 		if (!skb_pkt_type_ok(*ptype))
165 			return -EINVAL;
166 		flags |= SKBEDIT_F_PTYPE;
167 	}
168 
169 	if (tb[TCA_SKBEDIT_MARK] != NULL) {
170 		flags |= SKBEDIT_F_MARK;
171 		mark = nla_data(tb[TCA_SKBEDIT_MARK]);
172 	}
173 
174 	if (tb[TCA_SKBEDIT_MASK] != NULL) {
175 		flags |= SKBEDIT_F_MASK;
176 		mask = nla_data(tb[TCA_SKBEDIT_MASK]);
177 	}
178 
179 	if (tb[TCA_SKBEDIT_FLAGS] != NULL) {
180 		u64 *pure_flags = nla_data(tb[TCA_SKBEDIT_FLAGS]);
181 
182 		if (*pure_flags & SKBEDIT_F_TXQ_SKBHASH) {
183 			u16 *queue_mapping_max;
184 
185 			if (!tb[TCA_SKBEDIT_QUEUE_MAPPING] ||
186 			    !tb[TCA_SKBEDIT_QUEUE_MAPPING_MAX]) {
187 				NL_SET_ERR_MSG_MOD(extack, "Missing required range of queue_mapping.");
188 				return -EINVAL;
189 			}
190 
191 			queue_mapping_max =
192 				nla_data(tb[TCA_SKBEDIT_QUEUE_MAPPING_MAX]);
193 			if (*queue_mapping_max < *queue_mapping) {
194 				NL_SET_ERR_MSG_MOD(extack, "The range of queue_mapping is invalid, max < min.");
195 				return -EINVAL;
196 			}
197 
198 			mapping_mod = *queue_mapping_max - *queue_mapping + 1;
199 			flags |= SKBEDIT_F_TXQ_SKBHASH;
200 		}
201 		if (*pure_flags & SKBEDIT_F_INHERITDSFIELD)
202 			flags |= SKBEDIT_F_INHERITDSFIELD;
203 	}
204 
205 	parm = nla_data(tb[TCA_SKBEDIT_PARMS]);
206 	index = parm->index;
207 	err = tcf_idr_check_alloc(tn, &index, a, bind);
208 	if (err < 0)
209 		return err;
210 	exists = err;
211 	if (exists && bind)
212 		return 0;
213 
214 	if (!flags) {
215 		if (exists)
216 			tcf_idr_release(*a, bind);
217 		else
218 			tcf_idr_cleanup(tn, index);
219 		return -EINVAL;
220 	}
221 
222 	if (!exists) {
223 		ret = tcf_idr_create(tn, index, est, a,
224 				     &act_skbedit_ops, bind, true, act_flags);
225 		if (ret) {
226 			tcf_idr_cleanup(tn, index);
227 			return ret;
228 		}
229 
230 		d = to_skbedit(*a);
231 		ret = ACT_P_CREATED;
232 	} else {
233 		d = to_skbedit(*a);
234 		if (!(act_flags & TCA_ACT_FLAGS_REPLACE)) {
235 			tcf_idr_release(*a, bind);
236 			return -EEXIST;
237 		}
238 	}
239 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
240 	if (err < 0)
241 		goto release_idr;
242 
243 	params_new = kzalloc(sizeof(*params_new), GFP_KERNEL);
244 	if (unlikely(!params_new)) {
245 		err = -ENOMEM;
246 		goto put_chain;
247 	}
248 
249 	params_new->flags = flags;
250 	if (flags & SKBEDIT_F_PRIORITY)
251 		params_new->priority = *priority;
252 	if (flags & SKBEDIT_F_QUEUE_MAPPING) {
253 		params_new->queue_mapping = *queue_mapping;
254 		params_new->mapping_mod = mapping_mod;
255 	}
256 	if (flags & SKBEDIT_F_MARK)
257 		params_new->mark = *mark;
258 	if (flags & SKBEDIT_F_PTYPE)
259 		params_new->ptype = *ptype;
260 	/* default behaviour is to use all the bits */
261 	params_new->mask = 0xffffffff;
262 	if (flags & SKBEDIT_F_MASK)
263 		params_new->mask = *mask;
264 
265 	spin_lock_bh(&d->tcf_lock);
266 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
267 	params_new = rcu_replace_pointer(d->params, params_new,
268 					 lockdep_is_held(&d->tcf_lock));
269 	spin_unlock_bh(&d->tcf_lock);
270 	if (params_new)
271 		kfree_rcu(params_new, rcu);
272 	if (goto_ch)
273 		tcf_chain_put_by_act(goto_ch);
274 
275 	return ret;
276 put_chain:
277 	if (goto_ch)
278 		tcf_chain_put_by_act(goto_ch);
279 release_idr:
280 	tcf_idr_release(*a, bind);
281 	return err;
282 }
283 
284 static int tcf_skbedit_dump(struct sk_buff *skb, struct tc_action *a,
285 			    int bind, int ref)
286 {
287 	unsigned char *b = skb_tail_pointer(skb);
288 	struct tcf_skbedit *d = to_skbedit(a);
289 	struct tcf_skbedit_params *params;
290 	struct tc_skbedit opt = {
291 		.index   = d->tcf_index,
292 		.refcnt  = refcount_read(&d->tcf_refcnt) - ref,
293 		.bindcnt = atomic_read(&d->tcf_bindcnt) - bind,
294 	};
295 	u64 pure_flags = 0;
296 	struct tcf_t t;
297 
298 	spin_lock_bh(&d->tcf_lock);
299 	params = rcu_dereference_protected(d->params,
300 					   lockdep_is_held(&d->tcf_lock));
301 	opt.action = d->tcf_action;
302 
303 	if (nla_put(skb, TCA_SKBEDIT_PARMS, sizeof(opt), &opt))
304 		goto nla_put_failure;
305 	if ((params->flags & SKBEDIT_F_PRIORITY) &&
306 	    nla_put_u32(skb, TCA_SKBEDIT_PRIORITY, params->priority))
307 		goto nla_put_failure;
308 	if ((params->flags & SKBEDIT_F_QUEUE_MAPPING) &&
309 	    nla_put_u16(skb, TCA_SKBEDIT_QUEUE_MAPPING, params->queue_mapping))
310 		goto nla_put_failure;
311 	if ((params->flags & SKBEDIT_F_MARK) &&
312 	    nla_put_u32(skb, TCA_SKBEDIT_MARK, params->mark))
313 		goto nla_put_failure;
314 	if ((params->flags & SKBEDIT_F_PTYPE) &&
315 	    nla_put_u16(skb, TCA_SKBEDIT_PTYPE, params->ptype))
316 		goto nla_put_failure;
317 	if ((params->flags & SKBEDIT_F_MASK) &&
318 	    nla_put_u32(skb, TCA_SKBEDIT_MASK, params->mask))
319 		goto nla_put_failure;
320 	if (params->flags & SKBEDIT_F_INHERITDSFIELD)
321 		pure_flags |= SKBEDIT_F_INHERITDSFIELD;
322 	if (params->flags & SKBEDIT_F_TXQ_SKBHASH) {
323 		if (nla_put_u16(skb, TCA_SKBEDIT_QUEUE_MAPPING_MAX,
324 				params->queue_mapping + params->mapping_mod - 1))
325 			goto nla_put_failure;
326 
327 		pure_flags |= SKBEDIT_F_TXQ_SKBHASH;
328 	}
329 	if (pure_flags != 0 &&
330 	    nla_put(skb, TCA_SKBEDIT_FLAGS, sizeof(pure_flags), &pure_flags))
331 		goto nla_put_failure;
332 
333 	tcf_tm_dump(&t, &d->tcf_tm);
334 	if (nla_put_64bit(skb, TCA_SKBEDIT_TM, sizeof(t), &t, TCA_SKBEDIT_PAD))
335 		goto nla_put_failure;
336 	spin_unlock_bh(&d->tcf_lock);
337 
338 	return skb->len;
339 
340 nla_put_failure:
341 	spin_unlock_bh(&d->tcf_lock);
342 	nlmsg_trim(skb, b);
343 	return -1;
344 }
345 
346 static void tcf_skbedit_cleanup(struct tc_action *a)
347 {
348 	struct tcf_skbedit *d = to_skbedit(a);
349 	struct tcf_skbedit_params *params;
350 
351 	params = rcu_dereference_protected(d->params, 1);
352 	if (params)
353 		kfree_rcu(params, rcu);
354 }
355 
356 static size_t tcf_skbedit_get_fill_size(const struct tc_action *act)
357 {
358 	return nla_total_size(sizeof(struct tc_skbedit))
359 		+ nla_total_size(sizeof(u32)) /* TCA_SKBEDIT_PRIORITY */
360 		+ nla_total_size(sizeof(u16)) /* TCA_SKBEDIT_QUEUE_MAPPING */
361 		+ nla_total_size(sizeof(u16)) /* TCA_SKBEDIT_QUEUE_MAPPING_MAX */
362 		+ nla_total_size(sizeof(u32)) /* TCA_SKBEDIT_MARK */
363 		+ nla_total_size(sizeof(u16)) /* TCA_SKBEDIT_PTYPE */
364 		+ nla_total_size(sizeof(u32)) /* TCA_SKBEDIT_MASK */
365 		+ nla_total_size_64bit(sizeof(u64)); /* TCA_SKBEDIT_FLAGS */
366 }
367 
368 static int tcf_skbedit_offload_act_setup(struct tc_action *act, void *entry_data,
369 					 u32 *index_inc, bool bind,
370 					 struct netlink_ext_ack *extack)
371 {
372 	if (bind) {
373 		struct flow_action_entry *entry = entry_data;
374 
375 		if (is_tcf_skbedit_mark(act)) {
376 			entry->id = FLOW_ACTION_MARK;
377 			entry->mark = tcf_skbedit_mark(act);
378 		} else if (is_tcf_skbedit_ptype(act)) {
379 			entry->id = FLOW_ACTION_PTYPE;
380 			entry->ptype = tcf_skbedit_ptype(act);
381 		} else if (is_tcf_skbedit_priority(act)) {
382 			entry->id = FLOW_ACTION_PRIORITY;
383 			entry->priority = tcf_skbedit_priority(act);
384 		} else if (is_tcf_skbedit_tx_queue_mapping(act)) {
385 			NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"queue_mapping\" option is used on transmit side");
386 			return -EOPNOTSUPP;
387 		} else if (is_tcf_skbedit_rx_queue_mapping(act)) {
388 			entry->id = FLOW_ACTION_RX_QUEUE_MAPPING;
389 			entry->rx_queue = tcf_skbedit_rx_queue_mapping(act);
390 		} else if (is_tcf_skbedit_inheritdsfield(act)) {
391 			NL_SET_ERR_MSG_MOD(extack, "Offload not supported when \"inheritdsfield\" option is used");
392 			return -EOPNOTSUPP;
393 		} else {
394 			NL_SET_ERR_MSG_MOD(extack, "Unsupported skbedit option offload");
395 			return -EOPNOTSUPP;
396 		}
397 		*index_inc = 1;
398 	} else {
399 		struct flow_offload_action *fl_action = entry_data;
400 
401 		if (is_tcf_skbedit_mark(act))
402 			fl_action->id = FLOW_ACTION_MARK;
403 		else if (is_tcf_skbedit_ptype(act))
404 			fl_action->id = FLOW_ACTION_PTYPE;
405 		else if (is_tcf_skbedit_priority(act))
406 			fl_action->id = FLOW_ACTION_PRIORITY;
407 		else if (is_tcf_skbedit_rx_queue_mapping(act))
408 			fl_action->id = FLOW_ACTION_RX_QUEUE_MAPPING;
409 		else
410 			return -EOPNOTSUPP;
411 	}
412 
413 	return 0;
414 }
415 
416 static struct tc_action_ops act_skbedit_ops = {
417 	.kind		=	"skbedit",
418 	.id		=	TCA_ID_SKBEDIT,
419 	.owner		=	THIS_MODULE,
420 	.act		=	tcf_skbedit_act,
421 	.stats_update	=	tcf_skbedit_stats_update,
422 	.dump		=	tcf_skbedit_dump,
423 	.init		=	tcf_skbedit_init,
424 	.cleanup	=	tcf_skbedit_cleanup,
425 	.get_fill_size	=	tcf_skbedit_get_fill_size,
426 	.offload_act_setup =	tcf_skbedit_offload_act_setup,
427 	.size		=	sizeof(struct tcf_skbedit),
428 };
429 
430 static __net_init int skbedit_init_net(struct net *net)
431 {
432 	struct tc_action_net *tn = net_generic(net, act_skbedit_ops.net_id);
433 
434 	return tc_action_net_init(net, tn, &act_skbedit_ops);
435 }
436 
437 static void __net_exit skbedit_exit_net(struct list_head *net_list)
438 {
439 	tc_action_net_exit(net_list, act_skbedit_ops.net_id);
440 }
441 
442 static struct pernet_operations skbedit_net_ops = {
443 	.init = skbedit_init_net,
444 	.exit_batch = skbedit_exit_net,
445 	.id   = &act_skbedit_ops.net_id,
446 	.size = sizeof(struct tc_action_net),
447 };
448 
449 MODULE_AUTHOR("Alexander Duyck, <alexander.h.duyck@intel.com>");
450 MODULE_DESCRIPTION("SKB Editing");
451 MODULE_LICENSE("GPL");
452 
453 static int __init skbedit_init_module(void)
454 {
455 	return tcf_register_action(&act_skbedit_ops, &skbedit_net_ops);
456 }
457 
458 static void __exit skbedit_cleanup_module(void)
459 {
460 	tcf_unregister_action(&act_skbedit_ops, &skbedit_net_ops);
461 }
462 
463 module_init(skbedit_init_module);
464 module_exit(skbedit_cleanup_module);
465