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 #define SMC_MAX_HOSTNAME_LEN 32
33 #define SMC_MAX_EID_LEN 32
34
35 extern struct proto smc_proto;
36 extern struct proto smc_proto6;
37
38 #ifdef ATOMIC64_INIT
39 #define KERNEL_HAS_ATOMIC64
40 #endif
41
42 enum smc_state { /* possible states of an SMC socket */
43 SMC_ACTIVE = 1,
44 SMC_INIT = 2,
45 SMC_CLOSED = 7,
46 SMC_LISTEN = 10,
47 /* normal close */
48 SMC_PEERCLOSEWAIT1 = 20,
49 SMC_PEERCLOSEWAIT2 = 21,
50 SMC_APPFINCLOSEWAIT = 24,
51 SMC_APPCLOSEWAIT1 = 22,
52 SMC_APPCLOSEWAIT2 = 23,
53 SMC_PEERFINCLOSEWAIT = 25,
54 /* abnormal close */
55 SMC_PEERABORTWAIT = 26,
56 SMC_PROCESSABORT = 27,
57 };
58
59 struct smc_link_group;
60
61 struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */
62 u8 type;
63 } __aligned(1);
64
65 struct smc_cdc_conn_state_flags {
66 #if defined(__BIG_ENDIAN_BITFIELD)
67 u8 peer_done_writing : 1; /* Sending done indicator */
68 u8 peer_conn_closed : 1; /* Peer connection closed indicator */
69 u8 peer_conn_abort : 1; /* Abnormal close indicator */
70 u8 reserved : 5;
71 #elif defined(__LITTLE_ENDIAN_BITFIELD)
72 u8 reserved : 5;
73 u8 peer_conn_abort : 1;
74 u8 peer_conn_closed : 1;
75 u8 peer_done_writing : 1;
76 #endif
77 };
78
79 struct smc_cdc_producer_flags {
80 #if defined(__BIG_ENDIAN_BITFIELD)
81 u8 write_blocked : 1; /* Writing Blocked, no rx buf space */
82 u8 urg_data_pending : 1; /* Urgent Data Pending */
83 u8 urg_data_present : 1; /* Urgent Data Present */
84 u8 cons_curs_upd_req : 1; /* cursor update requested */
85 u8 failover_validation : 1;/* message replay due to failover */
86 u8 reserved : 3;
87 #elif defined(__LITTLE_ENDIAN_BITFIELD)
88 u8 reserved : 3;
89 u8 failover_validation : 1;
90 u8 cons_curs_upd_req : 1;
91 u8 urg_data_present : 1;
92 u8 urg_data_pending : 1;
93 u8 write_blocked : 1;
94 #endif
95 };
96
97 /* in host byte order */
98 union smc_host_cursor { /* SMC cursor - an offset in an RMBE */
99 struct {
100 u16 reserved;
101 u16 wrap; /* window wrap sequence number */
102 u32 count; /* cursor (= offset) part */
103 };
104 #ifdef KERNEL_HAS_ATOMIC64
105 atomic64_t acurs; /* for atomic processing */
106 #else
107 u64 acurs; /* for atomic processing */
108 #endif
109 } __aligned(8);
110
111 /* in host byte order, except for flag bitfields in network byte order */
112 struct smc_host_cdc_msg { /* Connection Data Control message */
113 struct smc_wr_rx_hdr common; /* .type = 0xFE */
114 u8 len; /* length = 44 */
115 u16 seqno; /* connection seq # */
116 u32 token; /* alert_token */
117 union smc_host_cursor prod; /* producer cursor */
118 union smc_host_cursor cons; /* consumer cursor,
119 * piggy backed "ack"
120 */
121 struct smc_cdc_producer_flags prod_flags; /* conn. tx/rx status */
122 struct smc_cdc_conn_state_flags conn_state_flags; /* peer conn. status*/
123 u8 reserved[18];
124 } __aligned(8);
125
126 enum smc_urg_state {
127 SMC_URG_VALID = 1, /* data present */
128 SMC_URG_NOTYET = 2, /* data pending */
129 SMC_URG_READ = 3, /* data was already read */
130 };
131
132 struct smc_mark_woken {
133 bool woken;
134 void *key;
135 wait_queue_entry_t wait_entry;
136 };
137
138 struct smc_connection {
139 struct rb_node alert_node;
140 struct smc_link_group *lgr; /* link group of connection */
141 struct smc_link *lnk; /* assigned SMC-R link */
142 u32 alert_token_local; /* unique conn. id */
143 u8 peer_rmbe_idx; /* from tcp handshake */
144 int peer_rmbe_size; /* size of peer rx buffer */
145 atomic_t peer_rmbe_space;/* remaining free bytes in peer
146 * rmbe
147 */
148 int rtoken_idx; /* idx to peer RMB rkey/addr */
149
150 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */
151 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */
152 int rmbe_size_short;/* compressed notation */
153 int rmbe_update_limit;
154 /* lower limit for consumer
155 * cursor update
156 */
157
158 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging
159 * buffer for CDC msg send
160 * .prod cf. TCP snd_nxt
161 * .cons cf. TCP sends ack
162 */
163 union smc_host_cursor local_tx_ctrl_fin;
164 /* prod crsr - confirmed by peer
165 */
166 union smc_host_cursor tx_curs_prep; /* tx - prepared data
167 * snd_max..wmem_alloc
168 */
169 union smc_host_cursor tx_curs_sent; /* tx - sent data
170 * snd_nxt ?
171 */
172 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer
173 * snd-wnd-begin ?
174 */
175 atomic_t sndbuf_space; /* remaining space in sndbuf */
176 u16 tx_cdc_seq; /* sequence # for CDC send */
177 u16 tx_cdc_seq_fin; /* sequence # - tx completed */
178 spinlock_t send_lock; /* protect wr_sends */
179 atomic_t cdc_pend_tx_wr; /* number of pending tx CDC wqe
180 * - inc when post wqe,
181 * - dec on polled tx cqe
182 */
183 wait_queue_head_t cdc_pend_tx_wq; /* wakeup on no cdc_pend_tx_wr*/
184 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */
185 u32 tx_off; /* base offset in peer rmb */
186
187 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl.
188 * .prod cf. TCP rcv_nxt
189 * .cons cf. TCP snd_una
190 */
191 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer
192 * source of snd_una ?
193 */
194 union smc_host_cursor urg_curs; /* points at urgent byte */
195 enum smc_urg_state urg_state;
196 bool urg_tx_pend; /* urgent data staged */
197 bool urg_rx_skip_pend;
198 /* indicate urgent oob data
199 * read, but previous regular
200 * data still pending
201 */
202 char urg_rx_byte; /* urgent byte */
203 atomic_t bytes_to_rcv; /* arrived data,
204 * not yet received
205 */
206 atomic_t splice_pending; /* number of spliced bytes
207 * pending processing
208 */
209 #ifndef KERNEL_HAS_ATOMIC64
210 spinlock_t acurs_lock; /* protect cursors */
211 #endif
212 struct work_struct close_work; /* peer sent some closing */
213 struct work_struct abort_work; /* abort the connection */
214 struct tasklet_struct rx_tsklet; /* Receiver tasklet for SMC-D */
215 u8 rx_off; /* receive offset:
216 * 0 for SMC-R, 32 for SMC-D
217 */
218 u64 peer_token; /* SMC-D token of peer */
219 u8 killed : 1; /* abnormal termination */
220 u8 out_of_sync : 1; /* out of sync with peer */
221 };
222
223 struct smc_sock { /* smc sock container */
224 struct sock sk;
225 struct socket *clcsock; /* internal tcp socket */
226 void (*clcsk_state_change)(struct sock *sk);
227 /* original stat_change fct. */
228 void (*clcsk_data_ready)(struct sock *sk);
229 /* original data_ready fct. */
230 void (*clcsk_write_space)(struct sock *sk);
231 /* original write_space fct. */
232 void (*clcsk_error_report)(struct sock *sk);
233 /* original error_report fct. */
234 struct smc_connection conn; /* smc connection */
235 struct smc_sock *listen_smc; /* listen parent */
236 struct work_struct connect_work; /* handle non-blocking connect*/
237 struct work_struct tcp_listen_work;/* handle tcp socket accepts */
238 struct work_struct smc_listen_work;/* prepare new accept socket */
239 struct list_head accept_q; /* sockets to be accepted */
240 spinlock_t accept_q_lock; /* protects accept_q */
241 bool use_fallback; /* fallback to tcp */
242 int fallback_rsn; /* reason for fallback */
243 u32 peer_diagnosis; /* decline reason from peer */
244 int sockopt_defer_accept;
245 /* sockopt TCP_DEFER_ACCEPT
246 * value
247 */
248 u8 wait_close_tx_prepared : 1;
249 /* shutdown wr or close
250 * started, waiting for unsent
251 * data to be sent
252 */
253 u8 connect_nonblock : 1;
254 /* non-blocking connect in
255 * flight
256 */
257 struct mutex clcsock_release_lock;
258 /* protects clcsock of a listen
259 * socket
260 * */
261 };
262
smc_sk(const struct sock * sk)263 static inline struct smc_sock *smc_sk(const struct sock *sk)
264 {
265 return (struct smc_sock *)sk;
266 }
267
smc_clcsock_user_data(struct sock * clcsk)268 static inline struct smc_sock *smc_clcsock_user_data(struct sock *clcsk)
269 {
270 return (struct smc_sock *)
271 ((uintptr_t)clcsk->sk_user_data & ~SK_USER_DATA_NOCOPY);
272 }
273
274 extern struct workqueue_struct *smc_hs_wq; /* wq for handshake work */
275 extern struct workqueue_struct *smc_close_wq; /* wq for close work */
276
277 #define SMC_SYSTEMID_LEN 8
278
279 extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */
280
281 #define ntohll(x) be64_to_cpu(x)
282 #define htonll(x) cpu_to_be64(x)
283
284 /* convert an u32 value into network byte order, store it into a 3 byte field */
hton24(u8 * net,u32 host)285 static inline void hton24(u8 *net, u32 host)
286 {
287 __be32 t;
288
289 t = cpu_to_be32(host);
290 memcpy(net, ((u8 *)&t) + 1, 3);
291 }
292
293 /* convert a received 3 byte field into host byte order*/
ntoh24(u8 * net)294 static inline u32 ntoh24(u8 *net)
295 {
296 __be32 t = 0;
297
298 memcpy(((u8 *)&t) + 1, net, 3);
299 return be32_to_cpu(t);
300 }
301
302 #ifdef CONFIG_XFRM
using_ipsec(struct smc_sock * smc)303 static inline bool using_ipsec(struct smc_sock *smc)
304 {
305 return (smc->clcsock->sk->sk_policy[0] ||
306 smc->clcsock->sk->sk_policy[1]) ? true : false;
307 }
308 #else
using_ipsec(struct smc_sock * smc)309 static inline bool using_ipsec(struct smc_sock *smc)
310 {
311 return false;
312 }
313 #endif
314
315 struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock);
316 void smc_close_non_accepted(struct sock *sk);
317
smc_sock_set_flag(struct sock * sk,enum sock_flags flag)318 static inline void smc_sock_set_flag(struct sock *sk, enum sock_flags flag)
319 {
320 set_bit(flag, &sk->sk_flags);
321 }
322
323 #endif /* __SMC_H */
324