1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * Definitions for the SMC module (socket related)
6 *
7 * Copyright IBM Corp. 2016
8 *
9 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
10 */
11 #ifndef __SMC_H
12 #define __SMC_H
13
14 #include <linux/socket.h>
15 #include <linux/types.h>
16 #include <linux/compiler.h> /* __aligned */
17 #include <net/sock.h>
18
19 #include "smc_ib.h"
20
21 #define SMC_V1 1 /* SMC version V1 */
22 #define SMC_V2 2 /* SMC version V2 */
23 #define SMC_RELEASE 0
24
25 #define SMCPROTO_SMC 0 /* SMC protocol, IPv4 */
26 #define SMCPROTO_SMC6 1 /* SMC protocol, IPv6 */
27
28 #define SMC_MAX_ISM_DEVS 8 /* max # of proposed non-native ISM
29 * devices
30 */
31
32 extern struct proto smc_proto;
33 extern struct proto smc_proto6;
34
35 #ifdef ATOMIC64_INIT
36 #define KERNEL_HAS_ATOMIC64
37 #endif
38
39 enum smc_state { /* possible states of an SMC socket */
40 SMC_ACTIVE = 1,
41 SMC_INIT = 2,
42 SMC_CLOSED = 7,
43 SMC_LISTEN = 10,
44 /* normal close */
45 SMC_PEERCLOSEWAIT1 = 20,
46 SMC_PEERCLOSEWAIT2 = 21,
47 SMC_APPFINCLOSEWAIT = 24,
48 SMC_APPCLOSEWAIT1 = 22,
49 SMC_APPCLOSEWAIT2 = 23,
50 SMC_PEERFINCLOSEWAIT = 25,
51 /* abnormal close */
52 SMC_PEERABORTWAIT = 26,
53 SMC_PROCESSABORT = 27,
54 };
55
56 struct smc_link_group;
57
58 struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */
59 union {
60 u8 type;
61 #if defined(__BIG_ENDIAN_BITFIELD)
62 struct {
63 u8 llc_version:4,
64 llc_type:4;
65 };
66 #elif defined(__LITTLE_ENDIAN_BITFIELD)
67 struct {
68 u8 llc_type:4,
69 llc_version:4;
70 };
71 #endif
72 };
73 } __aligned(1);
74
75 struct smc_cdc_conn_state_flags {
76 #if defined(__BIG_ENDIAN_BITFIELD)
77 u8 peer_done_writing : 1; /* Sending done indicator */
78 u8 peer_conn_closed : 1; /* Peer connection closed indicator */
79 u8 peer_conn_abort : 1; /* Abnormal close indicator */
80 u8 reserved : 5;
81 #elif defined(__LITTLE_ENDIAN_BITFIELD)
82 u8 reserved : 5;
83 u8 peer_conn_abort : 1;
84 u8 peer_conn_closed : 1;
85 u8 peer_done_writing : 1;
86 #endif
87 };
88
89 struct smc_cdc_producer_flags {
90 #if defined(__BIG_ENDIAN_BITFIELD)
91 u8 write_blocked : 1; /* Writing Blocked, no rx buf space */
92 u8 urg_data_pending : 1; /* Urgent Data Pending */
93 u8 urg_data_present : 1; /* Urgent Data Present */
94 u8 cons_curs_upd_req : 1; /* cursor update requested */
95 u8 failover_validation : 1;/* message replay due to failover */
96 u8 reserved : 3;
97 #elif defined(__LITTLE_ENDIAN_BITFIELD)
98 u8 reserved : 3;
99 u8 failover_validation : 1;
100 u8 cons_curs_upd_req : 1;
101 u8 urg_data_present : 1;
102 u8 urg_data_pending : 1;
103 u8 write_blocked : 1;
104 #endif
105 };
106
107 /* in host byte order */
108 union smc_host_cursor { /* SMC cursor - an offset in an RMBE */
109 struct {
110 u16 reserved;
111 u16 wrap; /* window wrap sequence number */
112 u32 count; /* cursor (= offset) part */
113 };
114 #ifdef KERNEL_HAS_ATOMIC64
115 atomic64_t acurs; /* for atomic processing */
116 #else
117 u64 acurs; /* for atomic processing */
118 #endif
119 } __aligned(8);
120
121 /* in host byte order, except for flag bitfields in network byte order */
122 struct smc_host_cdc_msg { /* Connection Data Control message */
123 struct smc_wr_rx_hdr common; /* .type = 0xFE */
124 u8 len; /* length = 44 */
125 u16 seqno; /* connection seq # */
126 u32 token; /* alert_token */
127 union smc_host_cursor prod; /* producer cursor */
128 union smc_host_cursor cons; /* consumer cursor,
129 * piggy backed "ack"
130 */
131 struct smc_cdc_producer_flags prod_flags; /* conn. tx/rx status */
132 struct smc_cdc_conn_state_flags conn_state_flags; /* peer conn. status*/
133 u8 reserved[18];
134 } __aligned(8);
135
136 enum smc_urg_state {
137 SMC_URG_VALID = 1, /* data present */
138 SMC_URG_NOTYET = 2, /* data pending */
139 SMC_URG_READ = 3, /* data was already read */
140 };
141
142 struct smc_connection {
143 struct rb_node alert_node;
144 struct smc_link_group *lgr; /* link group of connection */
145 struct smc_link *lnk; /* assigned SMC-R link */
146 u32 alert_token_local; /* unique conn. id */
147 u8 peer_rmbe_idx; /* from tcp handshake */
148 int peer_rmbe_size; /* size of peer rx buffer */
149 atomic_t peer_rmbe_space;/* remaining free bytes in peer
150 * rmbe
151 */
152 int rtoken_idx; /* idx to peer RMB rkey/addr */
153
154 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */
155 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */
156 int rmbe_size_short;/* compressed notation */
157 int rmbe_update_limit;
158 /* lower limit for consumer
159 * cursor update
160 */
161
162 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging
163 * buffer for CDC msg send
164 * .prod cf. TCP snd_nxt
165 * .cons cf. TCP sends ack
166 */
167 union smc_host_cursor local_tx_ctrl_fin;
168 /* prod crsr - confirmed by peer
169 */
170 union smc_host_cursor tx_curs_prep; /* tx - prepared data
171 * snd_max..wmem_alloc
172 */
173 union smc_host_cursor tx_curs_sent; /* tx - sent data
174 * snd_nxt ?
175 */
176 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer
177 * snd-wnd-begin ?
178 */
179 atomic_t sndbuf_space; /* remaining space in sndbuf */
180 u16 tx_cdc_seq; /* sequence # for CDC send */
181 u16 tx_cdc_seq_fin; /* sequence # - tx completed */
182 spinlock_t send_lock; /* protect wr_sends */
183 atomic_t cdc_pend_tx_wr; /* number of pending tx CDC wqe
184 * - inc when post wqe,
185 * - dec on polled tx cqe
186 */
187 wait_queue_head_t cdc_pend_tx_wq; /* wakeup on no cdc_pend_tx_wr*/
188 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */
189 u32 tx_off; /* base offset in peer rmb */
190
191 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl.
192 * .prod cf. TCP rcv_nxt
193 * .cons cf. TCP snd_una
194 */
195 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer
196 * source of snd_una ?
197 */
198 union smc_host_cursor urg_curs; /* points at urgent byte */
199 enum smc_urg_state urg_state;
200 bool urg_tx_pend; /* urgent data staged */
201 bool urg_rx_skip_pend;
202 /* indicate urgent oob data
203 * read, but previous regular
204 * data still pending
205 */
206 char urg_rx_byte; /* urgent byte */
207 atomic_t bytes_to_rcv; /* arrived data,
208 * not yet received
209 */
210 atomic_t splice_pending; /* number of spliced bytes
211 * pending processing
212 */
213 #ifndef KERNEL_HAS_ATOMIC64
214 spinlock_t acurs_lock; /* protect cursors */
215 #endif
216 struct work_struct close_work; /* peer sent some closing */
217 struct work_struct abort_work; /* abort the connection */
218 struct tasklet_struct rx_tsklet; /* Receiver tasklet for SMC-D */
219 u8 rx_off; /* receive offset:
220 * 0 for SMC-R, 32 for SMC-D
221 */
222 u64 peer_token; /* SMC-D token of peer */
223 u8 killed : 1; /* abnormal termination */
224 u8 out_of_sync : 1; /* out of sync with peer */
225 };
226
227 struct smc_sock { /* smc sock container */
228 struct sock sk;
229 struct socket *clcsock; /* internal tcp socket */
230 void (*clcsk_data_ready)(struct sock *sk);
231 /* original data_ready fct. **/
232 struct smc_connection conn; /* smc connection */
233 struct smc_sock *listen_smc; /* listen parent */
234 struct work_struct connect_work; /* handle non-blocking connect*/
235 struct work_struct tcp_listen_work;/* handle tcp socket accepts */
236 struct work_struct smc_listen_work;/* prepare new accept socket */
237 struct list_head accept_q; /* sockets to be accepted */
238 spinlock_t accept_q_lock; /* protects accept_q */
239 bool use_fallback; /* fallback to tcp */
240 int fallback_rsn; /* reason for fallback */
241 u32 peer_diagnosis; /* decline reason from peer */
242 int sockopt_defer_accept;
243 /* sockopt TCP_DEFER_ACCEPT
244 * value
245 */
246 u8 wait_close_tx_prepared : 1;
247 /* shutdown wr or close
248 * started, waiting for unsent
249 * data to be sent
250 */
251 u8 connect_nonblock : 1;
252 /* non-blocking connect in
253 * flight
254 */
255 struct mutex clcsock_release_lock;
256 /* protects clcsock of a listen
257 * socket
258 * */
259 };
260
smc_sk(const struct sock * sk)261 static inline struct smc_sock *smc_sk(const struct sock *sk)
262 {
263 return (struct smc_sock *)sk;
264 }
265
266 extern struct workqueue_struct *smc_hs_wq; /* wq for handshake work */
267 extern struct workqueue_struct *smc_close_wq; /* wq for close work */
268
269 #define SMC_SYSTEMID_LEN 8
270
271 extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */
272
273 #define ntohll(x) be64_to_cpu(x)
274 #define htonll(x) cpu_to_be64(x)
275
276 /* convert an u32 value into network byte order, store it into a 3 byte field */
hton24(u8 * net,u32 host)277 static inline void hton24(u8 *net, u32 host)
278 {
279 __be32 t;
280
281 t = cpu_to_be32(host);
282 memcpy(net, ((u8 *)&t) + 1, 3);
283 }
284
285 /* convert a received 3 byte field into host byte order*/
ntoh24(u8 * net)286 static inline u32 ntoh24(u8 *net)
287 {
288 __be32 t = 0;
289
290 memcpy(((u8 *)&t) + 1, net, 3);
291 return be32_to_cpu(t);
292 }
293
294 #ifdef CONFIG_XFRM
using_ipsec(struct smc_sock * smc)295 static inline bool using_ipsec(struct smc_sock *smc)
296 {
297 return (smc->clcsock->sk->sk_policy[0] ||
298 smc->clcsock->sk->sk_policy[1]) ? true : false;
299 }
300 #else
using_ipsec(struct smc_sock * smc)301 static inline bool using_ipsec(struct smc_sock *smc)
302 {
303 return false;
304 }
305 #endif
306
307 struct smc_gidlist;
308
309 struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock);
310 void smc_close_non_accepted(struct sock *sk);
311 void smc_fill_gid_list(struct smc_link_group *lgr,
312 struct smc_gidlist *gidlist,
313 struct smc_ib_device *known_dev, u8 *known_gid);
314
315 #endif /* __SMC_H */
316