xref: /openbmc/linux/net/ipv4/gre_offload.c (revision a59511d1)
1 /*
2  *	IPV4 GSO/GRO offload support
3  *	Linux INET implementation
4  *
5  *	This program is free software; you can redistribute it and/or
6  *	modify it under the terms of the GNU General Public License
7  *	as published by the Free Software Foundation; either version
8  *	2 of the License, or (at your option) any later version.
9  *
10  *	GRE GSO support
11  */
12 
13 #include <linux/skbuff.h>
14 #include <linux/init.h>
15 #include <net/protocol.h>
16 #include <net/gre.h>
17 
18 static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
19 				       netdev_features_t features)
20 {
21 	int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
22 	struct sk_buff *segs = ERR_PTR(-EINVAL);
23 	u16 mac_offset = skb->mac_header;
24 	__be16 protocol = skb->protocol;
25 	u16 mac_len = skb->mac_len;
26 	int gre_offset, outer_hlen;
27 	bool need_csum, ufo;
28 
29 	if (unlikely(skb_shinfo(skb)->gso_type &
30 				~(SKB_GSO_TCPV4 |
31 				  SKB_GSO_TCPV6 |
32 				  SKB_GSO_UDP |
33 				  SKB_GSO_DODGY |
34 				  SKB_GSO_TCP_ECN |
35 				  SKB_GSO_GRE |
36 				  SKB_GSO_GRE_CSUM |
37 				  SKB_GSO_IPIP |
38 				  SKB_GSO_SIT)))
39 		goto out;
40 
41 	if (!skb->encapsulation)
42 		goto out;
43 
44 	if (unlikely(tnl_hlen < sizeof(struct gre_base_hdr)))
45 		goto out;
46 
47 	if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
48 		goto out;
49 
50 	/* setup inner skb. */
51 	skb->encapsulation = 0;
52 	SKB_GSO_CB(skb)->encap_level = 0;
53 	__skb_pull(skb, tnl_hlen);
54 	skb_reset_mac_header(skb);
55 	skb_set_network_header(skb, skb_inner_network_offset(skb));
56 	skb->mac_len = skb_inner_network_offset(skb);
57 	skb->protocol = skb->inner_protocol;
58 
59 	need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_GRE_CSUM);
60 	skb->encap_hdr_csum = need_csum;
61 
62 	ufo = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
63 
64 	features &= skb->dev->hw_enc_features;
65 
66 	/* The only checksum offload we care about from here on out is the
67 	 * outer one so strip the existing checksum feature flags based
68 	 * on the fact that we will be computing our checksum in software.
69 	 */
70 	if (ufo) {
71 		features &= ~NETIF_F_CSUM_MASK;
72 		if (!need_csum)
73 			features |= NETIF_F_HW_CSUM;
74 	}
75 
76 	/* segment inner packet. */
77 	segs = skb_mac_gso_segment(skb, features);
78 	if (IS_ERR_OR_NULL(segs)) {
79 		skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
80 				     mac_len);
81 		goto out;
82 	}
83 
84 	outer_hlen = skb_tnl_header_len(skb);
85 	gre_offset = outer_hlen - tnl_hlen;
86 	skb = segs;
87 	do {
88 		struct gre_base_hdr *greh;
89 		__be32 *pcsum;
90 
91 		/* Set up inner headers if we are offloading inner checksum */
92 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
93 			skb_reset_inner_headers(skb);
94 			skb->encapsulation = 1;
95 		}
96 
97 		skb->mac_len = mac_len;
98 		skb->protocol = protocol;
99 
100 		__skb_push(skb, outer_hlen);
101 		skb_reset_mac_header(skb);
102 		skb_set_network_header(skb, mac_len);
103 		skb_set_transport_header(skb, gre_offset);
104 
105 		if (!need_csum)
106 			continue;
107 
108 		greh = (struct gre_base_hdr *)skb_transport_header(skb);
109 		pcsum = (__be32 *)(greh + 1);
110 
111 		*pcsum = 0;
112 		*(__sum16 *)pcsum = gso_make_checksum(skb, 0);
113 	} while ((skb = skb->next));
114 out:
115 	return segs;
116 }
117 
118 static struct sk_buff **gre_gro_receive(struct sk_buff **head,
119 					struct sk_buff *skb)
120 {
121 	struct sk_buff **pp = NULL;
122 	struct sk_buff *p;
123 	const struct gre_base_hdr *greh;
124 	unsigned int hlen, grehlen;
125 	unsigned int off;
126 	int flush = 1;
127 	struct packet_offload *ptype;
128 	__be16 type;
129 
130 	if (NAPI_GRO_CB(skb)->encap_mark)
131 		goto out;
132 
133 	NAPI_GRO_CB(skb)->encap_mark = 1;
134 
135 	off = skb_gro_offset(skb);
136 	hlen = off + sizeof(*greh);
137 	greh = skb_gro_header_fast(skb, off);
138 	if (skb_gro_header_hard(skb, hlen)) {
139 		greh = skb_gro_header_slow(skb, hlen, off);
140 		if (unlikely(!greh))
141 			goto out;
142 	}
143 
144 	/* Only support version 0 and K (key), C (csum) flags. Note that
145 	 * although the support for the S (seq#) flag can be added easily
146 	 * for GRO, this is problematic for GSO hence can not be enabled
147 	 * here because a GRO pkt may end up in the forwarding path, thus
148 	 * requiring GSO support to break it up correctly.
149 	 */
150 	if ((greh->flags & ~(GRE_KEY|GRE_CSUM)) != 0)
151 		goto out;
152 
153 	type = greh->protocol;
154 
155 	rcu_read_lock();
156 	ptype = gro_find_receive_by_type(type);
157 	if (!ptype)
158 		goto out_unlock;
159 
160 	grehlen = GRE_HEADER_SECTION;
161 
162 	if (greh->flags & GRE_KEY)
163 		grehlen += GRE_HEADER_SECTION;
164 
165 	if (greh->flags & GRE_CSUM)
166 		grehlen += GRE_HEADER_SECTION;
167 
168 	hlen = off + grehlen;
169 	if (skb_gro_header_hard(skb, hlen)) {
170 		greh = skb_gro_header_slow(skb, hlen, off);
171 		if (unlikely(!greh))
172 			goto out_unlock;
173 	}
174 
175 	/* Don't bother verifying checksum if we're going to flush anyway. */
176 	if ((greh->flags & GRE_CSUM) && !NAPI_GRO_CB(skb)->flush) {
177 		if (skb_gro_checksum_simple_validate(skb))
178 			goto out_unlock;
179 
180 		skb_gro_checksum_try_convert(skb, IPPROTO_GRE, 0,
181 					     null_compute_pseudo);
182 	}
183 
184 	for (p = *head; p; p = p->next) {
185 		const struct gre_base_hdr *greh2;
186 
187 		if (!NAPI_GRO_CB(p)->same_flow)
188 			continue;
189 
190 		/* The following checks are needed to ensure only pkts
191 		 * from the same tunnel are considered for aggregation.
192 		 * The criteria for "the same tunnel" includes:
193 		 * 1) same version (we only support version 0 here)
194 		 * 2) same protocol (we only support ETH_P_IP for now)
195 		 * 3) same set of flags
196 		 * 4) same key if the key field is present.
197 		 */
198 		greh2 = (struct gre_base_hdr *)(p->data + off);
199 
200 		if (greh2->flags != greh->flags ||
201 		    greh2->protocol != greh->protocol) {
202 			NAPI_GRO_CB(p)->same_flow = 0;
203 			continue;
204 		}
205 		if (greh->flags & GRE_KEY) {
206 			/* compare keys */
207 			if (*(__be32 *)(greh2+1) != *(__be32 *)(greh+1)) {
208 				NAPI_GRO_CB(p)->same_flow = 0;
209 				continue;
210 			}
211 		}
212 	}
213 
214 	skb_gro_pull(skb, grehlen);
215 
216 	/* Adjusted NAPI_GRO_CB(skb)->csum after skb_gro_pull()*/
217 	skb_gro_postpull_rcsum(skb, greh, grehlen);
218 
219 	pp = ptype->callbacks.gro_receive(head, skb);
220 	flush = 0;
221 
222 out_unlock:
223 	rcu_read_unlock();
224 out:
225 	NAPI_GRO_CB(skb)->flush |= flush;
226 
227 	return pp;
228 }
229 
230 static int gre_gro_complete(struct sk_buff *skb, int nhoff)
231 {
232 	struct gre_base_hdr *greh = (struct gre_base_hdr *)(skb->data + nhoff);
233 	struct packet_offload *ptype;
234 	unsigned int grehlen = sizeof(*greh);
235 	int err = -ENOENT;
236 	__be16 type;
237 
238 	skb->encapsulation = 1;
239 	skb_shinfo(skb)->gso_type = SKB_GSO_GRE;
240 
241 	type = greh->protocol;
242 	if (greh->flags & GRE_KEY)
243 		grehlen += GRE_HEADER_SECTION;
244 
245 	if (greh->flags & GRE_CSUM)
246 		grehlen += GRE_HEADER_SECTION;
247 
248 	rcu_read_lock();
249 	ptype = gro_find_complete_by_type(type);
250 	if (ptype)
251 		err = ptype->callbacks.gro_complete(skb, nhoff + grehlen);
252 
253 	rcu_read_unlock();
254 
255 	skb_set_inner_mac_header(skb, nhoff + grehlen);
256 
257 	return err;
258 }
259 
260 static const struct net_offload gre_offload = {
261 	.callbacks = {
262 		.gso_segment = gre_gso_segment,
263 		.gro_receive = gre_gro_receive,
264 		.gro_complete = gre_gro_complete,
265 	},
266 };
267 
268 static int __init gre_offload_init(void)
269 {
270 	return inet_add_offload(&gre_offload, IPPROTO_GRE);
271 }
272 device_initcall(gre_offload_init);
273