projects/linux·blob master · refs

360 lines · 9.5 KiB

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 */
19struct 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 ver field */
30#define MCTP_HDR_VER_MASK GENMASK(3, 0)
31
32/* Definitions for flags_seq_tag field */
33#define MCTP_HDR_FLAG_SOM BIT(7)
34#define MCTP_HDR_FLAG_EOM BIT(6)
35#define MCTP_HDR_FLAG_TO BIT(3)
36#define MCTP_HDR_FLAGS GENMASK(5, 3)
37#define MCTP_HDR_SEQ_SHIFT 4
38#define MCTP_HDR_SEQ_MASK GENMASK(1, 0)
39#define MCTP_HDR_TAG_SHIFT 0
40#define MCTP_HDR_TAG_MASK GENMASK(2, 0)
41
42#define MCTP_INITIAL_DEFAULT_NET 1
43
44static inline bool mctp_address_unicast(mctp_eid_t eid)
45{
46 return eid >= 8 && eid < 255;
47}
48
49static inline bool mctp_address_broadcast(mctp_eid_t eid)
50{
51 return eid == 255;
52}
53
54static inline bool mctp_address_null(mctp_eid_t eid)
55{
56 return eid == 0;
57}
58
59static inline bool mctp_address_matches(mctp_eid_t match, mctp_eid_t eid)
60{
61 return match == eid || match == MCTP_ADDR_ANY;
62}
63
64static inline struct mctp_hdr *mctp_hdr(struct sk_buff *skb)
65{
66 return (struct mctp_hdr *)skb_network_header(skb);
67}
68
69/* socket implementation */
70struct mctp_sock {
71 struct sock sk;
72
73 /* bind() params */
74 unsigned int bind_net;
75 mctp_eid_t bind_local_addr;
76 mctp_eid_t bind_peer_addr;
77 unsigned int bind_peer_net;
78 bool bind_peer_set;
79 __u8 bind_type;
80
81 /* sendmsg()/recvmsg() uses struct sockaddr_mctp_ext */
82 bool addr_ext;
83
84 /* list of mctp_sk_key, for incoming tag lookup. updates protected
85 * by sk->net->keys_lock
86 */
87 struct hlist_head keys;
88
89 /* mechanism for expiring allocated keys; will release an allocated
90 * tag, and any netdev state for a request/response pairing
91 */
92 struct timer_list key_expiry;
93};
94
95/* Key for matching incoming packets to sockets or reassembly contexts.
96 * Packets are matched on (peer EID, local EID, tag).
97 *
98 * Lifetime / locking requirements:
99 *
100 * - individual key data (ie, the struct itself) is protected by key->lock;
101 * changes must be made with that lock held.
102 *
103 * - the lookup fields: peer_addr, local_addr and tag are set before the
104 * key is added to lookup lists, and never updated.
105 *
106 * - A ref to the key must be held (throuh key->refs) if a pointer to the
107 * key is to be accessed after key->lock is released.
108 *
109 * - a mctp_sk_key contains a reference to a struct sock; this is valid
110 * for the life of the key. On sock destruction (through unhash), the key is
111 * removed from lists (see below), and marked invalid.
112 *
113 * - these mctp_sk_keys appear on two lists:
114 * 1) the struct mctp_sock->keys list
115 * 2) the struct netns_mctp->keys list
116 *
117 * presences on these lists requires a (single) refcount to be held; both
118 * lists are updated as a single operation.
119 *
120 * Updates and lookups in either list are performed under the
121 * netns_mctp->keys lock. Lookup functions will need to lock the key and
122 * take a reference before unlocking the keys_lock. Consequently, the list's
123 * keys_lock *cannot* be acquired with the individual key->lock held.
124 *
125 * - a key may have a sk_buff attached as part of an in-progress message
126 * reassembly (->reasm_head). The reasm data is protected by the individual
127 * key->lock.
128 *
129 * - there are two destruction paths for a mctp_sk_key:
130 *
131 * - through socket unhash (see mctp_sk_unhash). This performs the list
132 * removal under keys_lock.
133 *
134 * - where a key is established to receive a reply message: after receiving
135 * the (complete) reply, or during reassembly errors. Here, we clean up
136 * the reassembly context (marking reasm_dead, to prevent another from
137 * starting), and remove the socket from the netns & socket lists.
138 *
139 * - through an expiry timeout, on a per-socket timer
140 */
141struct mctp_sk_key {
142 unsigned int net;
143 mctp_eid_t peer_addr;
144 mctp_eid_t local_addr; /* MCTP_ADDR_ANY for local owned tags */
145 __u8 tag; /* incoming tag match; invert TO for local */
146
147 /* we hold a ref to sk when set */
148 struct sock *sk;
149
150 /* routing lookup list */
151 struct hlist_node hlist;
152
153 /* per-socket list */
154 struct hlist_node sklist;
155
156 /* lock protects against concurrent updates to the reassembly and
157 * expiry data below.
158 */
159 spinlock_t lock;
160
161 /* Keys are referenced during the output path, which may sleep */
162 refcount_t refs;
163
164 /* incoming fragment reassembly context */
165 struct sk_buff *reasm_head;
166 struct sk_buff **reasm_tailp;
167 bool reasm_dead;
168 u8 last_seq;
169
170 /* key validity */
171 bool valid;
172
173 /* expiry timeout; valid (above) cleared on expiry */
174 unsigned long expiry;
175
176 /* free to use for device flow state tracking. Initialised to
177 * zero on initial key creation
178 */
179 unsigned long dev_flow_state;
180 struct mctp_dev *dev;
181
182 /* a tag allocated with SIOCMCTPALLOCTAG ioctl will not expire
183 * automatically on timeout or response, instead SIOCMCTPDROPTAG
184 * is used.
185 */
186 bool manual_alloc;
187};
188
189struct mctp_skb_cb {
190 unsigned int magic;
191 unsigned int net;
192 /* fields below provide extended addressing for ingress to recvmsg() */
193 int ifindex;
194 unsigned char halen;
195 unsigned char haddr[MAX_ADDR_LEN];
196};
197
198/* skb control-block accessors with a little extra debugging for initial
199 * development.
200 *
201 * TODO: remove checks & mctp_skb_cb->magic; replace callers of __mctp_cb
202 * with mctp_cb().
203 *
204 * __mctp_cb() is only for the initial ingress code; we should see ->magic set
205 * at all times after this.
206 */
207static inline struct mctp_skb_cb *__mctp_cb(struct sk_buff *skb)
208{
209 struct mctp_skb_cb *cb = (void *)skb->cb;
210
211 cb->magic = 0x4d435450;
212 return cb;
213}
214
215static inline struct mctp_skb_cb *mctp_cb(struct sk_buff *skb)
216{
217 struct mctp_skb_cb *cb = (void *)skb->cb;
218
219 BUILD_BUG_ON(sizeof(struct mctp_skb_cb) > sizeof(skb->cb));
220 WARN_ON(cb->magic != 0x4d435450);
221 return cb;
222}
223
224/* If CONFIG_MCTP_FLOWS, we may add one of these as a SKB extension,
225 * indicating the flow to the device driver.
226 */
227struct mctp_flow {
228 struct mctp_sk_key *key;
229};
230
231struct mctp_dst;
232
233/* Route definition.
234 *
235 * These are held in the pernet->mctp.routes list, with RCU protection for
236 * removed routes. We hold a reference to the netdev; routes need to be
237 * dropped on NETDEV_UNREGISTER events.
238 *
239 * Updates to the route table are performed under rtnl; all reads under RCU,
240 * so routes cannot be referenced over a RCU grace period.
241 */
242struct mctp_route {
243 mctp_eid_t min, max;
244
245 unsigned char type;
246
247 unsigned int mtu;
248
249 enum {
250 MCTP_ROUTE_DIRECT,
251 MCTP_ROUTE_GATEWAY,
252 } dst_type;
253 union {
254 struct mctp_dev *dev;
255 struct mctp_fq_addr gateway;
256 };
257
258 int (*output)(struct mctp_dst *dst,
259 struct sk_buff *skb);
260
261 struct list_head list;
262 refcount_t refs;
263 struct rcu_head rcu;
264};
265
266/* Route lookup result: dst. Represents the results of a routing decision,
267 * but is only held over the individual routing operation.
268 *
269 * Will typically be stored on the caller stack, and must be released after
270 * usage.
271 */
272struct mctp_dst {
273 struct mctp_dev *dev;
274 unsigned int mtu;
275 mctp_eid_t nexthop;
276 mctp_eid_t saddr;
277
278 /* set for direct addressing */
279 unsigned char halen;
280 unsigned char haddr[MAX_ADDR_LEN];
281
282 int (*output)(struct mctp_dst *dst, struct sk_buff *skb);
283};
284
285int mctp_dst_from_extaddr(struct mctp_dst *dst, struct net *net, int ifindex,
286 unsigned char halen, const unsigned char *haddr);
287
288/* route interfaces */
289int mctp_route_lookup(struct net *net, unsigned int dnet,
290 mctp_eid_t daddr, struct mctp_dst *dst);
291
292void mctp_dst_release(struct mctp_dst *dst);
293
294/* always takes ownership of skb */
295int mctp_local_output(struct sock *sk, struct mctp_dst *dst,
296 struct sk_buff *skb, mctp_eid_t daddr, u8 req_tag);
297
298void mctp_key_unref(struct mctp_sk_key *key);
299struct mctp_sk_key *mctp_alloc_local_tag(struct mctp_sock *msk,
300 unsigned int netid,
301 mctp_eid_t local, mctp_eid_t peer,
302 bool manual, u8 *tagp);
303
304/* routing <--> device interface */
305unsigned int mctp_default_net(struct net *net);
306int mctp_default_net_set(struct net *net, unsigned int index);
307int mctp_route_add_local(struct mctp_dev *mdev, mctp_eid_t addr);
308int mctp_route_remove_local(struct mctp_dev *mdev, mctp_eid_t addr);
309void mctp_route_remove_dev(struct mctp_dev *mdev);
310
311/* neighbour definitions */
312enum mctp_neigh_source {
313 MCTP_NEIGH_STATIC,
314 MCTP_NEIGH_DISCOVER,
315};
316
317struct mctp_neigh {
318 struct mctp_dev *dev;
319 mctp_eid_t eid;
320 enum mctp_neigh_source source;
321
322 unsigned char ha[MAX_ADDR_LEN];
323
324 struct list_head list;
325 struct rcu_head rcu;
326};
327
328int mctp_neigh_init(void);
329void mctp_neigh_exit(void);
330
331// ret_hwaddr may be NULL, otherwise must have space for MAX_ADDR_LEN
332int mctp_neigh_lookup(struct mctp_dev *dev, mctp_eid_t eid,
333 void *ret_hwaddr);
334void mctp_neigh_remove_dev(struct mctp_dev *mdev);
335
336int mctp_routes_init(void);
337void mctp_routes_exit(void);
338
339int mctp_device_init(void);
340void mctp_device_exit(void);
341
342/* MCTP IDs and Codes from DMTF specification
343 * "DSP0239 Management Component Transport Protocol (MCTP) IDs and Codes"
344 * https://www.dmtf.org/sites/default/files/standards/documents/DSP0239_1.11.1.pdf
345 */
346enum mctp_phys_binding {
347 MCTP_PHYS_BINDING_UNSPEC = 0x00,
348 MCTP_PHYS_BINDING_SMBUS = 0x01,
349 MCTP_PHYS_BINDING_PCIE_VDM = 0x02,
350 MCTP_PHYS_BINDING_USB = 0x03,
351 MCTP_PHYS_BINDING_KCS = 0x04,
352 MCTP_PHYS_BINDING_SERIAL = 0x05,
353 MCTP_PHYS_BINDING_I3C = 0x06,
354 MCTP_PHYS_BINDING_MMBI = 0x07,
355 MCTP_PHYS_BINDING_PCC = 0x08,
356 MCTP_PHYS_BINDING_UCIE = 0x09,
357 MCTP_PHYS_BINDING_VENDOR = 0xFF,
358};
359
360#endif /* __NET_MCTP_H */