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_INITIAL_DEFAULT_NET 1 |
40 | |
41 | static inline bool mctp_address_unicast(mctp_eid_t eid) |
42 | { |
43 | return eid >= 8 && eid < 255; |
44 | } |
45 | |
46 | static inline bool mctp_address_broadcast(mctp_eid_t eid) |
47 | { |
48 | return eid == 255; |
49 | } |
50 | |
51 | static inline bool mctp_address_null(mctp_eid_t eid) |
52 | { |
53 | return eid == 0; |
54 | } |
55 | |
56 | static inline bool mctp_address_matches(mctp_eid_t match, mctp_eid_t eid) |
57 | { |
58 | return match == eid || match == MCTP_ADDR_ANY; |
59 | } |
60 | |
61 | static inline struct mctp_hdr *mctp_hdr(struct sk_buff *skb) |
62 | { |
63 | return (struct mctp_hdr *)skb_network_header(skb); |
64 | } |
65 | |
66 | /* socket implementation */ |
67 | struct mctp_sock { |
68 | struct sock sk; |
69 | |
70 | /* bind() params */ |
71 | unsigned int bind_net; |
72 | mctp_eid_t bind_addr; |
73 | __u8 bind_type; |
74 | |
75 | /* sendmsg()/recvmsg() uses struct sockaddr_mctp_ext */ |
76 | bool addr_ext; |
77 | |
78 | /* list of mctp_sk_key, for incoming tag lookup. updates protected |
79 | * by sk->net->keys_lock |
80 | */ |
81 | struct hlist_head keys; |
82 | |
83 | /* mechanism for expiring allocated keys; will release an allocated |
84 | * tag, and any netdev state for a request/response pairing |
85 | */ |
86 | struct timer_list key_expiry; |
87 | }; |
88 | |
89 | /* Key for matching incoming packets to sockets or reassembly contexts. |
90 | * Packets are matched on (peer EID, local EID, tag). |
91 | * |
92 | * Lifetime / locking requirements: |
93 | * |
94 | * - individual key data (ie, the struct itself) is protected by key->lock; |
95 | * changes must be made with that lock held. |
96 | * |
97 | * - the lookup fields: peer_addr, local_addr and tag are set before the |
98 | * key is added to lookup lists, and never updated. |
99 | * |
100 | * - A ref to the key must be held (throuh key->refs) if a pointer to the |
101 | * key is to be accessed after key->lock is released. |
102 | * |
103 | * - a mctp_sk_key contains a reference to a struct sock; this is valid |
104 | * for the life of the key. On sock destruction (through unhash), the key is |
105 | * removed from lists (see below), and marked invalid. |
106 | * |
107 | * - these mctp_sk_keys appear on two lists: |
108 | * 1) the struct mctp_sock->keys list |
109 | * 2) the struct netns_mctp->keys list |
110 | * |
111 | * presences on these lists requires a (single) refcount to be held; both |
112 | * lists are updated as a single operation. |
113 | * |
114 | * Updates and lookups in either list are performed under the |
115 | * netns_mctp->keys lock. Lookup functions will need to lock the key and |
116 | * take a reference before unlocking the keys_lock. Consequently, the list's |
117 | * keys_lock *cannot* be acquired with the individual key->lock held. |
118 | * |
119 | * - a key may have a sk_buff attached as part of an in-progress message |
120 | * reassembly (->reasm_head). The reasm data is protected by the individual |
121 | * key->lock. |
122 | * |
123 | * - there are two destruction paths for a mctp_sk_key: |
124 | * |
125 | * - through socket unhash (see mctp_sk_unhash). This performs the list |
126 | * removal under keys_lock. |
127 | * |
128 | * - where a key is established to receive a reply message: after receiving |
129 | * the (complete) reply, or during reassembly errors. Here, we clean up |
130 | * the reassembly context (marking reasm_dead, to prevent another from |
131 | * starting), and remove the socket from the netns & socket lists. |
132 | * |
133 | * - through an expiry timeout, on a per-socket timer |
134 | */ |
135 | struct mctp_sk_key { |
136 | unsigned int net; |
137 | mctp_eid_t peer_addr; |
138 | mctp_eid_t local_addr; /* MCTP_ADDR_ANY for local owned tags */ |
139 | __u8 tag; /* incoming tag match; invert TO for local */ |
140 | |
141 | /* we hold a ref to sk when set */ |
142 | struct sock *sk; |
143 | |
144 | /* routing lookup list */ |
145 | struct hlist_node hlist; |
146 | |
147 | /* per-socket list */ |
148 | struct hlist_node sklist; |
149 | |
150 | /* lock protects against concurrent updates to the reassembly and |
151 | * expiry data below. |
152 | */ |
153 | spinlock_t lock; |
154 | |
155 | /* Keys are referenced during the output path, which may sleep */ |
156 | refcount_t refs; |
157 | |
158 | /* incoming fragment reassembly context */ |
159 | struct sk_buff *reasm_head; |
160 | struct sk_buff **reasm_tailp; |
161 | bool reasm_dead; |
162 | u8 last_seq; |
163 | |
164 | /* key validity */ |
165 | bool valid; |
166 | |
167 | /* expiry timeout; valid (above) cleared on expiry */ |
168 | unsigned long expiry; |
169 | |
170 | /* free to use for device flow state tracking. Initialised to |
171 | * zero on initial key creation |
172 | */ |
173 | unsigned long dev_flow_state; |
174 | struct mctp_dev *dev; |
175 | |
176 | /* a tag allocated with SIOCMCTPALLOCTAG ioctl will not expire |
177 | * automatically on timeout or response, instead SIOCMCTPDROPTAG |
178 | * is used. |
179 | */ |
180 | bool manual_alloc; |
181 | }; |
182 | |
183 | struct mctp_skb_cb { |
184 | unsigned int magic; |
185 | unsigned int net; |
186 | int ifindex; /* extended/direct addressing if set */ |
187 | mctp_eid_t src; |
188 | unsigned char halen; |
189 | unsigned char haddr[MAX_ADDR_LEN]; |
190 | }; |
191 | |
192 | /* skb control-block accessors with a little extra debugging for initial |
193 | * development. |
194 | * |
195 | * TODO: remove checks & mctp_skb_cb->magic; replace callers of __mctp_cb |
196 | * with mctp_cb(). |
197 | * |
198 | * __mctp_cb() is only for the initial ingress code; we should see ->magic set |
199 | * at all times after this. |
200 | */ |
201 | static inline struct mctp_skb_cb *__mctp_cb(struct sk_buff *skb) |
202 | { |
203 | struct mctp_skb_cb *cb = (void *)skb->cb; |
204 | |
205 | cb->magic = 0x4d435450; |
206 | return cb; |
207 | } |
208 | |
209 | static inline struct mctp_skb_cb *mctp_cb(struct sk_buff *skb) |
210 | { |
211 | struct mctp_skb_cb *cb = (void *)skb->cb; |
212 | |
213 | BUILD_BUG_ON(sizeof(struct mctp_skb_cb) > sizeof(skb->cb)); |
214 | WARN_ON(cb->magic != 0x4d435450); |
215 | return (void *)(skb->cb); |
216 | } |
217 | |
218 | /* If CONFIG_MCTP_FLOWS, we may add one of these as a SKB extension, |
219 | * indicating the flow to the device driver. |
220 | */ |
221 | struct mctp_flow { |
222 | struct mctp_sk_key *key; |
223 | }; |
224 | |
225 | /* Route definition. |
226 | * |
227 | * These are held in the pernet->mctp.routes list, with RCU protection for |
228 | * removed routes. We hold a reference to the netdev; routes need to be |
229 | * dropped on NETDEV_UNREGISTER events. |
230 | * |
231 | * Updates to the route table are performed under rtnl; all reads under RCU, |
232 | * so routes cannot be referenced over a RCU grace period. Specifically: A |
233 | * caller cannot block between mctp_route_lookup and mctp_route_release() |
234 | */ |
235 | struct mctp_route { |
236 | mctp_eid_t min, max; |
237 | |
238 | unsigned char type; |
239 | unsigned int mtu; |
240 | struct mctp_dev *dev; |
241 | int (*output)(struct mctp_route *route, |
242 | struct sk_buff *skb); |
243 | |
244 | struct list_head list; |
245 | refcount_t refs; |
246 | struct rcu_head rcu; |
247 | }; |
248 | |
249 | /* route interfaces */ |
250 | struct mctp_route *mctp_route_lookup(struct net *net, unsigned int dnet, |
251 | mctp_eid_t daddr); |
252 | |
253 | /* always takes ownership of skb */ |
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 | unsigned int netid, |
260 | mctp_eid_t local, mctp_eid_t peer, |
261 | bool manual, u8 *tagp); |
262 | |
263 | /* routing <--> device interface */ |
264 | unsigned int mctp_default_net(struct net *net); |
265 | int mctp_default_net_set(struct net *net, unsigned int index); |
266 | int mctp_route_add_local(struct mctp_dev *mdev, mctp_eid_t addr); |
267 | int mctp_route_remove_local(struct mctp_dev *mdev, mctp_eid_t addr); |
268 | void mctp_route_remove_dev(struct mctp_dev *mdev); |
269 | |
270 | /* neighbour definitions */ |
271 | enum mctp_neigh_source { |
272 | MCTP_NEIGH_STATIC, |
273 | MCTP_NEIGH_DISCOVER, |
274 | }; |
275 | |
276 | struct mctp_neigh { |
277 | struct mctp_dev *dev; |
278 | mctp_eid_t eid; |
279 | enum mctp_neigh_source source; |
280 | |
281 | unsigned char ha[MAX_ADDR_LEN]; |
282 | |
283 | struct list_head list; |
284 | struct rcu_head rcu; |
285 | }; |
286 | |
287 | int mctp_neigh_init(void); |
288 | void mctp_neigh_exit(void); |
289 | |
290 | // ret_hwaddr may be NULL, otherwise must have space for MAX_ADDR_LEN |
291 | int mctp_neigh_lookup(struct mctp_dev *dev, mctp_eid_t eid, |
292 | void *ret_hwaddr); |
293 | void mctp_neigh_remove_dev(struct mctp_dev *mdev); |
294 | |
295 | int mctp_routes_init(void); |
296 | void mctp_routes_exit(void); |
297 | |
298 | void mctp_device_init(void); |
299 | void mctp_device_exit(void); |
300 | |
301 | #endif /* __NET_MCTP_H */ |
302 | |