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