xref: /openbmc/linux/include/net/mctp.h (revision 8dd3cdea)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Management Component Transport Protocol (MCTP)
4  *
5  * Copyright (c) 2021 Code Construct
6  * Copyright (c) 2021 Google
7  */
8 
9 #ifndef __NET_MCTP_H
10 #define __NET_MCTP_H
11 
12 #include <linux/bits.h>
13 #include <linux/mctp.h>
14 #include <linux/netdevice.h>
15 #include <net/net_namespace.h>
16 #include <net/sock.h>
17 
18 /* MCTP packet definitions */
19 struct mctp_hdr {
20 	u8	ver;
21 	u8	dest;
22 	u8	src;
23 	u8	flags_seq_tag;
24 };
25 
26 #define MCTP_VER_MIN	1
27 #define MCTP_VER_MAX	1
28 
29 /* Definitions for flags_seq_tag field */
30 #define MCTP_HDR_FLAG_SOM	BIT(7)
31 #define MCTP_HDR_FLAG_EOM	BIT(6)
32 #define MCTP_HDR_FLAG_TO	BIT(3)
33 #define MCTP_HDR_FLAGS		GENMASK(5, 3)
34 #define MCTP_HDR_SEQ_SHIFT	4
35 #define MCTP_HDR_SEQ_MASK	GENMASK(1, 0)
36 #define MCTP_HDR_TAG_SHIFT	0
37 #define MCTP_HDR_TAG_MASK	GENMASK(2, 0)
38 
39 #define MCTP_HEADER_MAXLEN	4
40 
41 #define MCTP_INITIAL_DEFAULT_NET	1
42 
43 static inline bool mctp_address_unicast(mctp_eid_t eid)
44 {
45 	return eid >= 8 && eid < 255;
46 }
47 
48 static inline bool mctp_address_broadcast(mctp_eid_t eid)
49 {
50 	return eid == 255;
51 }
52 
53 static inline bool mctp_address_null(mctp_eid_t eid)
54 {
55 	return eid == 0;
56 }
57 
58 static inline bool mctp_address_matches(mctp_eid_t match, mctp_eid_t eid)
59 {
60 	return match == eid || match == MCTP_ADDR_ANY;
61 }
62 
63 static inline struct mctp_hdr *mctp_hdr(struct sk_buff *skb)
64 {
65 	return (struct mctp_hdr *)skb_network_header(skb);
66 }
67 
68 /* socket implementation */
69 struct mctp_sock {
70 	struct sock	sk;
71 
72 	/* bind() params */
73 	unsigned int	bind_net;
74 	mctp_eid_t	bind_addr;
75 	__u8		bind_type;
76 
77 	/* sendmsg()/recvmsg() uses struct sockaddr_mctp_ext */
78 	bool		addr_ext;
79 
80 	/* list of mctp_sk_key, for incoming tag lookup. updates protected
81 	 * by sk->net->keys_lock
82 	 */
83 	struct hlist_head keys;
84 
85 	/* mechanism for expiring allocated keys; will release an allocated
86 	 * tag, and any netdev state for a request/response pairing
87 	 */
88 	struct timer_list key_expiry;
89 };
90 
91 /* Key for matching incoming packets to sockets or reassembly contexts.
92  * Packets are matched on (src,dest,tag).
93  *
94  * Lifetime / locking requirements:
95  *
96  *  - individual key data (ie, the struct itself) is protected by key->lock;
97  *    changes must be made with that lock held.
98  *
99  *  - the lookup fields: peer_addr, local_addr and tag are set before the
100  *    key is added to lookup lists, and never updated.
101  *
102  *  - A ref to the key must be held (throuh key->refs) if a pointer to the
103  *    key is to be accessed after key->lock is released.
104  *
105  *  - a mctp_sk_key contains a reference to a struct sock; this is valid
106  *    for the life of the key. On sock destruction (through unhash), the key is
107  *    removed from lists (see below), and marked invalid.
108  *
109  * - these mctp_sk_keys appear on two lists:
110  *     1) the struct mctp_sock->keys list
111  *     2) the struct netns_mctp->keys list
112  *
113  *   presences on these lists requires a (single) refcount to be held; both
114  *   lists are updated as a single operation.
115  *
116  *   Updates and lookups in either list are performed under the
117  *   netns_mctp->keys lock. Lookup functions will need to lock the key and
118  *   take a reference before unlocking the keys_lock. Consequently, the list's
119  *   keys_lock *cannot* be acquired with the individual key->lock held.
120  *
121  * - a key may have a sk_buff attached as part of an in-progress message
122  *   reassembly (->reasm_head). The reasm data is protected by the individual
123  *   key->lock.
124  *
125  * - there are two destruction paths for a mctp_sk_key:
126  *
127  *    - through socket unhash (see mctp_sk_unhash). This performs the list
128  *      removal under keys_lock.
129  *
130  *    - where a key is established to receive a reply message: after receiving
131  *      the (complete) reply, or during reassembly errors. Here, we clean up
132  *      the reassembly context (marking reasm_dead, to prevent another from
133  *      starting), and remove the socket from the netns & socket lists.
134  *
135  *    - through an expiry timeout, on a per-socket timer
136  */
137 struct mctp_sk_key {
138 	mctp_eid_t	peer_addr;
139 	mctp_eid_t	local_addr; /* MCTP_ADDR_ANY for local owned tags */
140 	__u8		tag; /* incoming tag match; invert TO for local */
141 
142 	/* we hold a ref to sk when set */
143 	struct sock	*sk;
144 
145 	/* routing lookup list */
146 	struct hlist_node hlist;
147 
148 	/* per-socket list */
149 	struct hlist_node sklist;
150 
151 	/* lock protects against concurrent updates to the reassembly and
152 	 * expiry data below.
153 	 */
154 	spinlock_t	lock;
155 
156 	/* Keys are referenced during the output path, which may sleep */
157 	refcount_t	refs;
158 
159 	/* incoming fragment reassembly context */
160 	struct sk_buff	*reasm_head;
161 	struct sk_buff	**reasm_tailp;
162 	bool		reasm_dead;
163 	u8		last_seq;
164 
165 	/* key validity */
166 	bool		valid;
167 
168 	/* expiry timeout; valid (above) cleared on expiry */
169 	unsigned long	expiry;
170 
171 	/* free to use for device flow state tracking. Initialised to
172 	 * zero on initial key creation
173 	 */
174 	unsigned long	dev_flow_state;
175 	struct mctp_dev	*dev;
176 
177 	/* a tag allocated with SIOCMCTPALLOCTAG ioctl will not expire
178 	 * automatically on timeout or response, instead SIOCMCTPDROPTAG
179 	 * is used.
180 	 */
181 	bool		manual_alloc;
182 };
183 
184 struct mctp_skb_cb {
185 	unsigned int	magic;
186 	unsigned int	net;
187 	int		ifindex; /* extended/direct addressing if set */
188 	mctp_eid_t	src;
189 	unsigned char	halen;
190 	unsigned char	haddr[MAX_ADDR_LEN];
191 };
192 
193 /* skb control-block accessors with a little extra debugging for initial
194  * development.
195  *
196  * TODO: remove checks & mctp_skb_cb->magic; replace callers of __mctp_cb
197  * with mctp_cb().
198  *
199  * __mctp_cb() is only for the initial ingress code; we should see ->magic set
200  * at all times after this.
201  */
202 static inline struct mctp_skb_cb *__mctp_cb(struct sk_buff *skb)
203 {
204 	struct mctp_skb_cb *cb = (void *)skb->cb;
205 
206 	cb->magic = 0x4d435450;
207 	return cb;
208 }
209 
210 static inline struct mctp_skb_cb *mctp_cb(struct sk_buff *skb)
211 {
212 	struct mctp_skb_cb *cb = (void *)skb->cb;
213 
214 	BUILD_BUG_ON(sizeof(struct mctp_skb_cb) > sizeof(skb->cb));
215 	WARN_ON(cb->magic != 0x4d435450);
216 	return (void *)(skb->cb);
217 }
218 
219 /* If CONFIG_MCTP_FLOWS, we may add one of these as a SKB extension,
220  * indicating the flow to the device driver.
221  */
222 struct mctp_flow {
223 	struct mctp_sk_key *key;
224 };
225 
226 /* Route definition.
227  *
228  * These are held in the pernet->mctp.routes list, with RCU protection for
229  * removed routes. We hold a reference to the netdev; routes need to be
230  * dropped on NETDEV_UNREGISTER events.
231  *
232  * Updates to the route table are performed under rtnl; all reads under RCU,
233  * so routes cannot be referenced over a RCU grace period. Specifically: A
234  * caller cannot block between mctp_route_lookup and mctp_route_release()
235  */
236 struct mctp_route {
237 	mctp_eid_t		min, max;
238 
239 	struct mctp_dev		*dev;
240 	unsigned int		mtu;
241 	unsigned char		type;
242 	int			(*output)(struct mctp_route *route,
243 					  struct sk_buff *skb);
244 
245 	struct list_head	list;
246 	refcount_t		refs;
247 	struct rcu_head		rcu;
248 };
249 
250 /* route interfaces */
251 struct mctp_route *mctp_route_lookup(struct net *net, unsigned int dnet,
252 				     mctp_eid_t daddr);
253 
254 int mctp_local_output(struct sock *sk, struct mctp_route *rt,
255 		      struct sk_buff *skb, mctp_eid_t daddr, u8 req_tag);
256 
257 void mctp_key_unref(struct mctp_sk_key *key);
258 struct mctp_sk_key *mctp_alloc_local_tag(struct mctp_sock *msk,
259 					 mctp_eid_t daddr, mctp_eid_t saddr,
260 					 bool manual, u8 *tagp);
261 
262 /* routing <--> device interface */
263 unsigned int mctp_default_net(struct net *net);
264 int mctp_default_net_set(struct net *net, unsigned int index);
265 int mctp_route_add_local(struct mctp_dev *mdev, mctp_eid_t addr);
266 int mctp_route_remove_local(struct mctp_dev *mdev, mctp_eid_t addr);
267 void mctp_route_remove_dev(struct mctp_dev *mdev);
268 
269 /* neighbour definitions */
270 enum mctp_neigh_source {
271 	MCTP_NEIGH_STATIC,
272 	MCTP_NEIGH_DISCOVER,
273 };
274 
275 struct mctp_neigh {
276 	struct mctp_dev		*dev;
277 	mctp_eid_t		eid;
278 	enum mctp_neigh_source	source;
279 
280 	unsigned char		ha[MAX_ADDR_LEN];
281 
282 	struct list_head	list;
283 	struct rcu_head		rcu;
284 };
285 
286 int mctp_neigh_init(void);
287 void mctp_neigh_exit(void);
288 
289 // ret_hwaddr may be NULL, otherwise must have space for MAX_ADDR_LEN
290 int mctp_neigh_lookup(struct mctp_dev *dev, mctp_eid_t eid,
291 		      void *ret_hwaddr);
292 void mctp_neigh_remove_dev(struct mctp_dev *mdev);
293 
294 int mctp_routes_init(void);
295 void mctp_routes_exit(void);
296 
297 void mctp_device_init(void);
298 void mctp_device_exit(void);
299 
300 #endif /* __NET_MCTP_H */
301