xref: /linux/net/sched/act_skbedit.c (revision f788bb2fb915a79eae0a160580766ba52984d61f)
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 int tcf_skbedit_act(struct sk_buff *skb, const struct tc_action *a,
27 			   struct tcf_result *res)
28 {
29 	struct tcf_skbedit *d = to_skbedit(a);
30 	struct tcf_skbedit_params *params;
31 	int action;
32 
33 	tcf_lastuse_update(&d->tcf_tm);
34 	bstats_cpu_update(this_cpu_ptr(d->common.cpu_bstats), skb);
35 
36 	params = rcu_dereference_bh(d->params);
37 	action = READ_ONCE(d->tcf_action);
38 
39 	if (params->flags & SKBEDIT_F_PRIORITY)
40 		skb->priority = params->priority;
41 	if (params->flags & SKBEDIT_F_INHERITDSFIELD) {
42 		int wlen = skb_network_offset(skb);
43 
44 		switch (skb_protocol(skb, true)) {
45 		case htons(ETH_P_IP):
46 			wlen += sizeof(struct iphdr);
47 			if (!pskb_may_pull(skb, wlen))
48 				goto err;
49 			skb->priority = ipv4_get_dsfield(ip_hdr(skb)) >> 2;
50 			break;
51 
52 		case htons(ETH_P_IPV6):
53 			wlen += sizeof(struct ipv6hdr);
54 			if (!pskb_may_pull(skb, wlen))
55 				goto err;
56 			skb->priority = ipv6_get_dsfield(ipv6_hdr(skb)) >> 2;
57 			break;
58 		}
59 	}
60 	if (params->flags & SKBEDIT_F_QUEUE_MAPPING &&
61 	    skb->dev->real_num_tx_queues > params->queue_mapping)
62 		skb_set_queue_mapping(skb, params->queue_mapping);
63 	if (params->flags & SKBEDIT_F_MARK) {
64 		skb->mark &= ~params->mask;
65 		skb->mark |= params->mark & params->mask;
66 	}
67 	if (params->flags & SKBEDIT_F_PTYPE)
68 		skb->pkt_type = params->ptype;
69 	return action;
70 
71 err:
72 	qstats_drop_inc(this_cpu_ptr(d->common.cpu_qstats));
73 	return TC_ACT_SHOT;
74 }
75 
76 static void tcf_skbedit_stats_update(struct tc_action *a, u64 bytes,
77 				     u32 packets, u64 lastuse, bool hw)
78 {
79 	struct tcf_skbedit *d = to_skbedit(a);
80 	struct tcf_t *tm = &d->tcf_tm;
81 
82 	tcf_action_update_stats(a, bytes, packets, false, hw);
83 	tm->lastuse = max_t(u64, tm->lastuse, lastuse);
84 }
85 
86 static const struct nla_policy skbedit_policy[TCA_SKBEDIT_MAX + 1] = {
87 	[TCA_SKBEDIT_PARMS]		= { .len = sizeof(struct tc_skbedit) },
88 	[TCA_SKBEDIT_PRIORITY]		= { .len = sizeof(u32) },
89 	[TCA_SKBEDIT_QUEUE_MAPPING]	= { .len = sizeof(u16) },
90 	[TCA_SKBEDIT_MARK]		= { .len = sizeof(u32) },
91 	[TCA_SKBEDIT_PTYPE]		= { .len = sizeof(u16) },
92 	[TCA_SKBEDIT_MASK]		= { .len = sizeof(u32) },
93 	[TCA_SKBEDIT_FLAGS]		= { .len = sizeof(u64) },
94 };
95 
96 static int tcf_skbedit_init(struct net *net, struct nlattr *nla,
97 			    struct nlattr *est, struct tc_action **a,
98 			    int ovr, int bind, bool rtnl_held,
99 			    struct tcf_proto *tp, u32 act_flags,
100 			    struct netlink_ext_ack *extack)
101 {
102 	struct tc_action_net *tn = net_generic(net, skbedit_net_id);
103 	struct tcf_skbedit_params *params_new;
104 	struct nlattr *tb[TCA_SKBEDIT_MAX + 1];
105 	struct tcf_chain *goto_ch = NULL;
106 	struct tc_skbedit *parm;
107 	struct tcf_skbedit *d;
108 	u32 flags = 0, *priority = NULL, *mark = NULL, *mask = NULL;
109 	u16 *queue_mapping = NULL, *ptype = NULL;
110 	bool exists = false;
111 	int ret = 0, err;
112 	u32 index;
113 
114 	if (nla == NULL)
115 		return -EINVAL;
116 
117 	err = nla_parse_nested_deprecated(tb, TCA_SKBEDIT_MAX, nla,
118 					  skbedit_policy, NULL);
119 	if (err < 0)
120 		return err;
121 
122 	if (tb[TCA_SKBEDIT_PARMS] == NULL)
123 		return -EINVAL;
124 
125 	if (tb[TCA_SKBEDIT_PRIORITY] != NULL) {
126 		flags |= SKBEDIT_F_PRIORITY;
127 		priority = nla_data(tb[TCA_SKBEDIT_PRIORITY]);
128 	}
129 
130 	if (tb[TCA_SKBEDIT_QUEUE_MAPPING] != NULL) {
131 		flags |= SKBEDIT_F_QUEUE_MAPPING;
132 		queue_mapping = nla_data(tb[TCA_SKBEDIT_QUEUE_MAPPING]);
133 	}
134 
135 	if (tb[TCA_SKBEDIT_PTYPE] != NULL) {
136 		ptype = nla_data(tb[TCA_SKBEDIT_PTYPE]);
137 		if (!skb_pkt_type_ok(*ptype))
138 			return -EINVAL;
139 		flags |= SKBEDIT_F_PTYPE;
140 	}
141 
142 	if (tb[TCA_SKBEDIT_MARK] != NULL) {
143 		flags |= SKBEDIT_F_MARK;
144 		mark = nla_data(tb[TCA_SKBEDIT_MARK]);
145 	}
146 
147 	if (tb[TCA_SKBEDIT_MASK] != NULL) {
148 		flags |= SKBEDIT_F_MASK;
149 		mask = nla_data(tb[TCA_SKBEDIT_MASK]);
150 	}
151 
152 	if (tb[TCA_SKBEDIT_FLAGS] != NULL) {
153 		u64 *pure_flags = nla_data(tb[TCA_SKBEDIT_FLAGS]);
154 
155 		if (*pure_flags & SKBEDIT_F_INHERITDSFIELD)
156 			flags |= SKBEDIT_F_INHERITDSFIELD;
157 	}
158 
159 	parm = nla_data(tb[TCA_SKBEDIT_PARMS]);
160 	index = parm->index;
161 	err = tcf_idr_check_alloc(tn, &index, a, bind);
162 	if (err < 0)
163 		return err;
164 	exists = err;
165 	if (exists && bind)
166 		return 0;
167 
168 	if (!flags) {
169 		if (exists)
170 			tcf_idr_release(*a, bind);
171 		else
172 			tcf_idr_cleanup(tn, index);
173 		return -EINVAL;
174 	}
175 
176 	if (!exists) {
177 		ret = tcf_idr_create(tn, index, est, a,
178 				     &act_skbedit_ops, bind, true, 0);
179 		if (ret) {
180 			tcf_idr_cleanup(tn, index);
181 			return ret;
182 		}
183 
184 		d = to_skbedit(*a);
185 		ret = ACT_P_CREATED;
186 	} else {
187 		d = to_skbedit(*a);
188 		if (!ovr) {
189 			tcf_idr_release(*a, bind);
190 			return -EEXIST;
191 		}
192 	}
193 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
194 	if (err < 0)
195 		goto release_idr;
196 
197 	params_new = kzalloc(sizeof(*params_new), GFP_KERNEL);
198 	if (unlikely(!params_new)) {
199 		err = -ENOMEM;
200 		goto put_chain;
201 	}
202 
203 	params_new->flags = flags;
204 	if (flags & SKBEDIT_F_PRIORITY)
205 		params_new->priority = *priority;
206 	if (flags & SKBEDIT_F_QUEUE_MAPPING)
207 		params_new->queue_mapping = *queue_mapping;
208 	if (flags & SKBEDIT_F_MARK)
209 		params_new->mark = *mark;
210 	if (flags & SKBEDIT_F_PTYPE)
211 		params_new->ptype = *ptype;
212 	/* default behaviour is to use all the bits */
213 	params_new->mask = 0xffffffff;
214 	if (flags & SKBEDIT_F_MASK)
215 		params_new->mask = *mask;
216 
217 	spin_lock_bh(&d->tcf_lock);
218 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
219 	params_new = rcu_replace_pointer(d->params, params_new,
220 					 lockdep_is_held(&d->tcf_lock));
221 	spin_unlock_bh(&d->tcf_lock);
222 	if (params_new)
223 		kfree_rcu(params_new, rcu);
224 	if (goto_ch)
225 		tcf_chain_put_by_act(goto_ch);
226 
227 	if (ret == ACT_P_CREATED)
228 		tcf_idr_insert(tn, *a);
229 	return ret;
230 put_chain:
231 	if (goto_ch)
232 		tcf_chain_put_by_act(goto_ch);
233 release_idr:
234 	tcf_idr_release(*a, bind);
235 	return err;
236 }
237 
238 static int tcf_skbedit_dump(struct sk_buff *skb, struct tc_action *a,
239 			    int bind, int ref)
240 {
241 	unsigned char *b = skb_tail_pointer(skb);
242 	struct tcf_skbedit *d = to_skbedit(a);
243 	struct tcf_skbedit_params *params;
244 	struct tc_skbedit opt = {
245 		.index   = d->tcf_index,
246 		.refcnt  = refcount_read(&d->tcf_refcnt) - ref,
247 		.bindcnt = atomic_read(&d->tcf_bindcnt) - bind,
248 	};
249 	u64 pure_flags = 0;
250 	struct tcf_t t;
251 
252 	spin_lock_bh(&d->tcf_lock);
253 	params = rcu_dereference_protected(d->params,
254 					   lockdep_is_held(&d->tcf_lock));
255 	opt.action = d->tcf_action;
256 
257 	if (nla_put(skb, TCA_SKBEDIT_PARMS, sizeof(opt), &opt))
258 		goto nla_put_failure;
259 	if ((params->flags & SKBEDIT_F_PRIORITY) &&
260 	    nla_put_u32(skb, TCA_SKBEDIT_PRIORITY, params->priority))
261 		goto nla_put_failure;
262 	if ((params->flags & SKBEDIT_F_QUEUE_MAPPING) &&
263 	    nla_put_u16(skb, TCA_SKBEDIT_QUEUE_MAPPING, params->queue_mapping))
264 		goto nla_put_failure;
265 	if ((params->flags & SKBEDIT_F_MARK) &&
266 	    nla_put_u32(skb, TCA_SKBEDIT_MARK, params->mark))
267 		goto nla_put_failure;
268 	if ((params->flags & SKBEDIT_F_PTYPE) &&
269 	    nla_put_u16(skb, TCA_SKBEDIT_PTYPE, params->ptype))
270 		goto nla_put_failure;
271 	if ((params->flags & SKBEDIT_F_MASK) &&
272 	    nla_put_u32(skb, TCA_SKBEDIT_MASK, params->mask))
273 		goto nla_put_failure;
274 	if (params->flags & SKBEDIT_F_INHERITDSFIELD)
275 		pure_flags |= SKBEDIT_F_INHERITDSFIELD;
276 	if (pure_flags != 0 &&
277 	    nla_put(skb, TCA_SKBEDIT_FLAGS, sizeof(pure_flags), &pure_flags))
278 		goto nla_put_failure;
279 
280 	tcf_tm_dump(&t, &d->tcf_tm);
281 	if (nla_put_64bit(skb, TCA_SKBEDIT_TM, sizeof(t), &t, TCA_SKBEDIT_PAD))
282 		goto nla_put_failure;
283 	spin_unlock_bh(&d->tcf_lock);
284 
285 	return skb->len;
286 
287 nla_put_failure:
288 	spin_unlock_bh(&d->tcf_lock);
289 	nlmsg_trim(skb, b);
290 	return -1;
291 }
292 
293 static void tcf_skbedit_cleanup(struct tc_action *a)
294 {
295 	struct tcf_skbedit *d = to_skbedit(a);
296 	struct tcf_skbedit_params *params;
297 
298 	params = rcu_dereference_protected(d->params, 1);
299 	if (params)
300 		kfree_rcu(params, rcu);
301 }
302 
303 static int tcf_skbedit_walker(struct net *net, struct sk_buff *skb,
304 			      struct netlink_callback *cb, int type,
305 			      const struct tc_action_ops *ops,
306 			      struct netlink_ext_ack *extack)
307 {
308 	struct tc_action_net *tn = net_generic(net, skbedit_net_id);
309 
310 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
311 }
312 
313 static int tcf_skbedit_search(struct net *net, struct tc_action **a, u32 index)
314 {
315 	struct tc_action_net *tn = net_generic(net, skbedit_net_id);
316 
317 	return tcf_idr_search(tn, a, index);
318 }
319 
320 static size_t tcf_skbedit_get_fill_size(const struct tc_action *act)
321 {
322 	return nla_total_size(sizeof(struct tc_skbedit))
323 		+ nla_total_size(sizeof(u32)) /* TCA_SKBEDIT_PRIORITY */
324 		+ nla_total_size(sizeof(u16)) /* TCA_SKBEDIT_QUEUE_MAPPING */
325 		+ nla_total_size(sizeof(u32)) /* TCA_SKBEDIT_MARK */
326 		+ nla_total_size(sizeof(u16)) /* TCA_SKBEDIT_PTYPE */
327 		+ nla_total_size(sizeof(u32)) /* TCA_SKBEDIT_MASK */
328 		+ nla_total_size_64bit(sizeof(u64)); /* TCA_SKBEDIT_FLAGS */
329 }
330 
331 static struct tc_action_ops act_skbedit_ops = {
332 	.kind		=	"skbedit",
333 	.id		=	TCA_ID_SKBEDIT,
334 	.owner		=	THIS_MODULE,
335 	.act		=	tcf_skbedit_act,
336 	.stats_update	=	tcf_skbedit_stats_update,
337 	.dump		=	tcf_skbedit_dump,
338 	.init		=	tcf_skbedit_init,
339 	.cleanup	=	tcf_skbedit_cleanup,
340 	.walk		=	tcf_skbedit_walker,
341 	.get_fill_size	=	tcf_skbedit_get_fill_size,
342 	.lookup		=	tcf_skbedit_search,
343 	.size		=	sizeof(struct tcf_skbedit),
344 };
345 
346 static __net_init int skbedit_init_net(struct net *net)
347 {
348 	struct tc_action_net *tn = net_generic(net, skbedit_net_id);
349 
350 	return tc_action_net_init(net, tn, &act_skbedit_ops);
351 }
352 
353 static void __net_exit skbedit_exit_net(struct list_head *net_list)
354 {
355 	tc_action_net_exit(net_list, skbedit_net_id);
356 }
357 
358 static struct pernet_operations skbedit_net_ops = {
359 	.init = skbedit_init_net,
360 	.exit_batch = skbedit_exit_net,
361 	.id   = &skbedit_net_id,
362 	.size = sizeof(struct tc_action_net),
363 };
364 
365 MODULE_AUTHOR("Alexander Duyck, <alexander.h.duyck@intel.com>");
366 MODULE_DESCRIPTION("SKB Editing");
367 MODULE_LICENSE("GPL");
368 
369 static int __init skbedit_init_module(void)
370 {
371 	return tcf_register_action(&act_skbedit_ops, &skbedit_net_ops);
372 }
373 
374 static void __exit skbedit_cleanup_module(void)
375 {
376 	tcf_unregister_action(&act_skbedit_ops, &skbedit_net_ops);
377 }
378 
379 module_init(skbedit_init_module);
380 module_exit(skbedit_cleanup_module);
381